Managed only Proprietary

Eden AI Managed marketplace

Name: Eden AI
Author: Eden AI

Eden AI is a managed marketplace: an OpenAI-compatible API in front of 1,104 models from 33–68 providers. There is no token markup or per-seat fee; the paid surface is a 5.5% credit-purchase fee. It cannot be self-hosted. Zero data retention and SOC 2 are published; a HIPAA BAA is not. You can point it at your own provider accounts. Beyond chat it also serves embeddings, image generation and audio.

Fields verified September 2, 2026 to September 25, 2026 · 54 dated entries · 68 source references

France company — outside US jurisdiction · EU region available

Compare with others Estimate cost Visit site ↗

Access at a glance

Whether this product can work for you at all, before features matter: who pays the model bill, where it can run, and which of your existing API calls keep working. Every value is the vendor’s own claim, linked to the page it came from.

Aggregator gateway, and deliberately broader than LLMs. The vendor calls it "a unified AI gateway that gives you access to 500+ AI models from 50+ providers through a single API" with "fallback, routing, and cost tracking built in" (AI gateway overview). The scope caveat matters for this catalogue: half the product is non-LLM. Alongside /v3/chat/completions sits /v3/universal-ai for "expert models" - OCR and document parsing (invoices, IDs, resumes, tables), speech-to-text and text-to-speech, translation, image analysis and generation, video generation, moderation and NER (LLMs vs Expert Models). The live feature catalog returned 7 families and 36 subfeatures on 2026-09-02, including a web family (search, map, research) that is not even mentioned in the docs (GET /v3/info). Read Eden AI as a multi-modality AI-API marketplace that also routes LLMs, not as an LLM-routing-first gateway.

Who pays the model bill

Your keys or their credits

You can start on their credits and move to your own provider accounts later.

You can spend Eden AI credits, or register your own provider keys in the dashboard so "requests routed to that provider will use your key and be billed directly by the provider", and the two can be mixed per provider (BYOK, observed 2026-09-02).

Merchant of record: Differs by mode. Credits mode: Eden AI invoices - you prepay credits by card, PayPal or bank transfer through Stripe and each request deducts its cost from the balance (Buying credits, Eden AI pricing). BYOK mode: the upstream provider bills you directly for requests routed to your own key (BYOK).

Key handling: Two key types. Your Eden AI tokens: a main account key plus "Custom API keys" created via /v2/user/custom_token/, each with its own type (api_token or sandbox_api_token), expiry date and balance/active_balance budget that stops the token once exhausted (Custom API keys). Upstream provider keys: BYOK credentials are entered in the dashboard per provider and used for requests routed there (BYOK). The DPA states encryption in transit and at rest (Data Processing Agreement); no KMS, secret-manager or key-rotation mechanics are documented.

Where it can run

1 of 5 shapes documented

Vendor-hosted
Self-host
Your VPC
On-premise
Air-gapped

Dimmed shapes are not documented by the vendor, which is not the same as unsupported.

hosted SaaS only: https://api.edenai.run/v3 with an EU-resident variant https://api.eu.edenai.run/v3/ (AI gateway overview, EU AI gateway). Self-host / hybrid VPC / on-prem / air-gapped: n.a. - not documented anywhere in the documentation index (docs index (llms.txt)). The Advanced plan lists "Private deployments" for "specific compliance needs" with no further detail, which is not a self-hosting statement (Eden AI pricing). Observed 2026-09-02.

Nothing to install: the vendor operates the gateway and its infrastructure "is hosted in secure European data centers" (Servers location). Legacy note for existing users - v3 is the current platform and pre-2026 accounts on old-app.edenai.run are "supported until the end of 2026", so migration is a dated obligation (AI gateway overview, page dateModified 2026-05-07).

API surfaces your code can keep using

4 of 7 documented, 2 partial

OpenAI chat POST /v1/chat/completions Yes

"Eden AI V3 provides full OpenAI API compatibility with multi-provider support. The endpoint follows OpenAI's exact format, making it a drop-in replacement" - POST https://api.edenai.run/v3/chat/completions, documented temperature, top_p, max_tokens, frequency_penalty, presence_penalty, stream, plus Eden-specific fallbacks and Claude thinking (Chat completions, observed 2026-09-02).
Anthropic messages POST /v1/messages Yes

A dedicated "Anthropic Messages API - native pass-through via litellm" is documented at POST https://api.edenai.run/v3/v1/messages, accepting Anthropic-shaped bodies plus Eden AI's fallbacks, router_candidates and cache_control, alongside a token-counting endpoint (Create Anthropic Message, docs index (llms.txt)).
OpenAI Responses POST /v1/responses Yes

"Eden AI supports the OpenAI Responses API - a stateful alternative to chat completions that stores conversation history server-side", with POST /v3/responses, GET/DELETE /v3/responses/{response_id}, store defaulting to true and previous_response_id chaining; it is also the documented path for web search on regular OpenAI models (Responses, Web search).
Embeddings POST /v1/embeddings Yes

OpenAI-compatible POST /v3/embeddings accepting arrays of inputs, with a companion catalog endpoint GET /v3/embeddings/models that needs no authentication and returned 29 embedding models from 9 providers on 2026-09-02 (Embeddings, live response from https://api.edenai.run/v3/embeddings/models).
Images POST /v1/images/generations Partly *

partial: there is no OpenAI /v1/images surface. Two other paths exist - image-capable LLMs return a base64 data URL through POST /v3/chat/completions with an image_config block (documented for Google Gemini Flash Image models only), and dedicated image APIs run through Universal AI as image/generation/... (Image generation, Create Async Job, LLMs vs Expert Models). The live catalog lists image subfeatures for generation, background removal, anonymization, object/face/logo detection, explicit-content, deepfake and AI-image detection (GET /v3/info).
Audio POST /v1/audio/* Partly *

partial: no OpenAI /v1/audio endpoint. Audio is an expert-model feature on Universal AI - text-to-speech (sync) and speech-to-text (async job with polling or webhook) - and the Universal AI endpoint explicitly does not stream (LLMs vs Expert Models, Webhooks, docs index (llms.txt)). The live feature catalog confirms the audio family with tts and speech_to_text_async only (GET /v3/info, 2026-09-02).
Batch jobs POST /v1/batches Not documented

not_documented as a batch surface. What exists is async single-job processing: POST /v3/universal-ai/async returns a public_id for polling or webhook delivery, one job per request (Create Async Job, Webhooks). No OpenAI-style /batches endpoint or bulk-file submission appears on the fetched pages or in the full documentation index (docs index (llms.txt)), even though the live catalog exposes a per-model supports_batch_inference capability flag (GET /v3/models).

An asterisk marks a qualified verdict: support that is indirect (SDK compatibility or provider passthrough rather than a native endpoint), or a gap that is narrower or wider than the label suggests. Read the note before porting.

Base URL https://api.edenai.run/v3 with Authorization: Bearer, split across two request shapes: POST /v3/chat/completions for LLMs (model string provider/model, OpenAI-compatible, SSE streaming) and POST /v3/universal-ai for expert models (model string feature/subfeature/provider[/model], unified status/cost/output envelope, no streaming, not OpenAI-SDK compatible) (AI gateway overview, LLMs vs Expert Models). Other surfaces: POST /v3/responses with GET/DELETE by id, POST /v3/embeddings plus an unauthenticated GET /v3/embeddings/models, POST /v3/moderations, POST /v3/upload for file storage, and POST https://api.edenai.run/v3/v1/messages documented as the "Anthropic Messages API - native pass-through via litellm" (Responses, Embeddings, Create Moderation, Create Anthropic Message). An EU-only host https://api.eu.edenai.run/v3/ mirrors the same paths (EU AI gateway). All observed 2026-09-02.

How much it reaches

Models 1,104

Upstream providers 33–68 vendor pages disagree

Models: Counted from the vendor’s own published list; no aggregate total is published.

Providers: The vendor publishes different totals on different pages; both bounds are shown.

Count of entries returned by the models API on 2026-09-25. Includes every entry exposed by that endpoint; not a count of unique base models.

Vendor figures disagree: "50+ providers" in the docs (AI gateway overview) versus "60+ AI providers" on the company page (About Eden AI), both observed 2026-09-02. Counted from the live catalog the same day: 33 distinct owned_by LLM providers (GET /v3/models) and 42 distinct expert-model providers (GET /v3/info), a union of 68 once overlaps such as OpenAI, Google, Microsoft, Amazon, Mistral and MiniMax are removed.

Whose models: All third-party: Eden AI hosts no models of its own and routes to provider APIs, returning the provider's owned_by and per-endpoint regions in its catalog. Counted 2026-09-02: 33 LLM serving providers (amazon, anthropic, azure, bytedance, cerebras, cloudflare, cohere, databricks, deepinfra, deepseek, fireworks_ai, flexai, google, groq, infomaniak, ionos, lilac, microsoft, minimax, mistral, moonshot, nebius, openai, ovhcloud, perplexityai, qwen, scaleway, tensorx, together_ai, vertex, xai, xiaomi, zai) (GET /v3/models) and 42 expert-model providers including affinda, api4ai, assembly, deepgram, deepl, elevenlabs, klippa, mindee, sightengine, stabilityai and veryfi (GET /v3/info).

Your own endpoints: No: no custom provider, custom base URL, self-hosted-model or vLLM/Ollama registration appears on any fetched page or in the complete documentation index (docs index (llms.txt), AI gateway overview). Model selection is limited to catalog entries or a provider-less model name handed to Eden AI's router (Provider routing). Checked 2026-09-02.

How it behaves in production

What happens when an upstream model is slow, wrong, or down — and what you can see and stop while it happens. Reliability features are recorded as where you configure them, not whether the vendor lists them, because almost every product here lists all of them.

2 of 6 reachable from code nothing documented on the request path no documented export

When something goes wrong

Each row says where the knob is, not whether the feature is on the marketing page. A control you can only reach by hand in someone else’s dashboard cannot be reviewed or version-controlled.

Request timeout Not documented

The vendor does not document this, so any behaviour you observe today is unversioned and may change.

For inference: no request-timeout parameter, header or default appears on the chat-completions, fallback, provider-routing or streaming pages (Chat completions, Fallback, Streaming). The only timeout published anywhere is on the webhook receiver side - 30 s per delivery attempt with up to 3 retries and exponential backoff capped at 30 s (Webhooks).
Retries Not documented

No retry count, backoff strategy or retry-on-status configuration appears on the fallback, provider-routing, chat-completions or rate-limit pages, and there is no retry page in the documentation index (Fallback, Provider routing, Chat completions, docs index (llms.txt)). The documented failure behaviour is fallback to the next candidate, not a retry counter. Default retry count: n.a. Backoff: n.a.
Fallback to another model Per request

Your application decides per call, so one noisy endpoint can have its own timeout without a redeploy.

ORDERED and per-request: a fallbacks array of up to 3 entries (a 4th returns 422), tried in order, on both /v3/chat/completions and Universal AI; entries can pin regions (model@eu, model@us) and duplicates are not de-duplicated. A second layer is provider-level failover for provider-less model names, switched off with routing.allow_fallbacks: false. "Failed attempts cost nothing", and x-edenai-metadata: enabled returns per-attempt provider, status and region (Fallback, Create Async Job, observed 2026-09-02).
Load balancing Per request

Weights are NOT user-settable. For a provider-less model name Eden AI spreads traffic across healthy endpoints with internal weighting and optimises for cost by default; your per-request controls are routing.sort (cost, speed, latency, exact, also usable as a model:latency suffix), routing.allowed_providers, routing.allow_fallbacks and sticky routing (on by default, to preserve provider prompt caches) (Provider routing). Smart routing goes further: model: "@edenai" lets Eden AI pick the model, optionally constrained by router_candidates (Smart routing).
Upstream health tracking Fixed, cannot change

The behaviour is fixed by the vendor. Predictable, but you cannot tune it for your workload.

Health is tracked by Eden AI, not by you. The router "ranks unhealthy providers last" while still keeping them as a last resort, and there is no interval, threshold, ejection window or circuit-breaker setting on the routing, fallback or rate-limit pages (Provider routing, Fallback). A public status page is published separately (Eden AI status).
Cross-region failover Not documented

Left empty because none of the existing enum values fits: region selection is real but it is per-request and per-endpoint, not a config file, and it is neither undocumented nor non-configurable. You can pin a region on each fallback entry (model@eu, model@us) (Fallback) or send the whole workload to the EU host https://api.eu.edenai.run/v3/, which exposes only EU-compatible providers and returns an error rather than routing outside the EU (EU AI gateway). 275 of 1,010 catalog endpoints carry an eu region tag, 378 us, 323 global, 36 ap, 1 ch (GET /v3/models, 2026-09-02).

Fallback chain: Ordered list — Try A, then B, then C. Simple and predictable, but every failover is all-or-nothing.

The reliability surface is deliberately thin and per-request: fallbacks (max 3), provider failover, region pinning and routing preferences are all body parameters, with no timeout, retry or circuit-breaker knobs at all (Fallback, Provider routing). Rate limiting is the sharpest operational edge, and the vendor's own numbers contradict: the plans table says "10 reqs/sec (+ on request)" while the prose on the same page says "7 requests/second by default, upgradable to 15 req/sec on request", and the dedicated page says "Every account starts at 10 requests per second", shared across all API keys with the organisation owner's limit acting as the ceiling for members (Plans & Pricing, Rate limits, both observed 2026-09-02).

How fast the hop is

Undisclosed vendor service

The vendor does not disclose what the request path runs on, so no overhead floor can be inferred at all.

vendor_saas: closed hosted service with no published source or runtime statement. The one architectural clue in the docs is the Anthropic endpoint, described as "native pass-through via litellm" (Create Anthropic Message, docs index (llms.txt)), and the catalog's capability/pricing field names (supports_prompt_caching, cache_read_input_token_cost) match LiteLLM's model-metadata vocabulary (GET /v3/models). No language, runtime or edge-network claim is made on the fetched pages.

You can run the request path yourself Not documented

Streaming Partly

No Docker image, Helm chart, binary, npm/PyPI package or Terraform artifact for a customer-run data plane appears on any fetched page or in the complete documentation index, which lists integrations (Claude Code, Cline, Codex CLI, Continue.dev, LangChain, LibreChat, n8n, Open WebUI, OpenAI SDKs, OpenClaw, OpenCode) but no deployment section (docs index (llms.txt), Eden AI pricing). The public GitHub organisation's 21 repositories are a docs site, a Claude Code skill, an n8n community node, an OpenClaw plugin, a cookbook and forks of third-party projects - no gateway server (GitHub org API, GitHub repos API, both queried 2026-09-02).

Streaming caveats: partial by endpoint. The LLM endpoint streams over SSE with stream: true, data: frames and a [DONE] marker, and stream_options.include_usage: true adds normalized cache usage and cost to the final usage event (Streaming, Prompt caching). The Universal AI expert-model endpoint is explicitly listed as not streaming, so OCR, speech and translation results arrive whole or as async jobs (LLMs vs Expert Models). Documented caveat on usage fields: "Fields that the provider does not report can be absent or zero" (Prompt caching).

This vendor publishes no latency or throughput figure for the routing layer. That is the most common case here, and it is why the architecture class above carries the comparison instead of a number.

No vendor or third-party benchmark exists to assess. The vendor's quantitative claims are availability and adoption figures (99.99% uptime, 200k+ developers), and independent coverage states neither has been audited (Founderland analysis, 2026-04-27); the public status page showed "All systems operational" with a 99.98% uptime figure on 2026-09-02 (Eden AI status).

What it will stop

nothing documented on the request path

No request-path policy controls are documented. That is not a fault in a product built purely for routing — but it means anything you need blocked has to be blocked before the call reaches here.

What you can see

No documented export

What gets logged Your choice

You decide whether bodies are captured, by setting or by header.

You can turn bodies off Yes

Yes - and the default is already opted out. Content storage is opt-in per project via the Log Retention toggle, so leaving it off (or switching it off) keeps prompts and outputs unstored; GDPR access, deletion and portability requests go to support@edenai.co (Data retention).

Traces Not documented

not_documented: no OpenTelemetry, OTLP, trace-id or span vocabulary appears on any fetched page or in the documentation index (Monitoring, docs index (llms.txt)). The nearest facility is per-request routing diagnostics - x-edenai-metadata: enabled returns each fallback attempt's provider, status and region - which is debugging metadata, not distributed tracing (Fallback).

configurable: an optional "Log Retention" toggle, set per project in the dashboard, switches on storage of full request and response content for debugging and audit; with it off, only metadata is retained (Data retention, observed 2026-09-02). No retention period is published for the opt-in mode.

Where telemetry can go

No documented export. Whatever this product records stays in its own interface, so it cannot become part of the monitoring you already run.

n.a. - no OTLP collector, Datadog/Grafana integration, Kafka or S3 sink, or CSV download is documented; usage and cost data leave only through the cost-monitoring REST endpoints (Monitoring, docs index (llms.txt)).

Records user feedback No

Scores live traffic No

n.a. - no feedback, rating, score or annotation endpoint appears on the monitoring page, in the API reference sections listed in the documentation index (chat, responses, embeddings, moderations, universal-ai, files, info, models, cost-monitoring, user-management) or on the pricing page (Monitoring, docs index (llms.txt)).

n.a. as a programmatic eval or scoring hook. The only comparison tooling named is a dashboard feature, "Compare models & performances", listed on the self-serve plan (Eden AI pricing); there is no evals page, dataset, scorer or offline-experiment API in the documentation index (docs index (llms.txt)).

Depends on the vendor’s SaaS: Yes by construction - there is no self-hosted mode, so cost monitoring, the model-comparison view and the log-retention toggle all live in Eden AI's dashboard and API (Monitoring, Eden AI pricing, Data retention).

Retention: There is no log-retention tier ladder: retention is a binary per-project toggle (metadata only, or full request/response content) with no stated duration for either mode, and no plan gating is mentioned (Data retention). Checked also on the plans and pricing pages, which do not mention log retention (Plans & Pricing, Eden AI pricing).

Whether it fits how you work

How much work stands between you and a first call, how different that is from running it in production, and whether it slots into the stack you already have. Recorded as the shape of the work rather than a number of minutes — how long it takes you depends on which accounts and quota you already hold, which no comparison can know.

to try: base url swap to run: base url swap fits 2 of 10 common stacks

Getting to a first call

2 numbered steps

Shape of the work Change one base URL

Your existing OpenAI-compatible client keeps working. You change a base URL and a key, and nothing else in your code moves.

Read off: the vendor’s own quickstart — 2 numbered steps.

The quickstart numbers only its two prerequisites (API token from the dashboard, then credits or a sandbox token); the call itself is an unnumbered code block, so "2" is the page's own count rather than a steps-to-first-call figure (First LLM Call).

Before step one

Your own provider key Optional

You can start on the product’s own credits and move to your own provider keys later.

. The first call needs only an Eden AI token plus credits (or a free sandbox token) - no provider account (First LLM Call). Bringing your own provider keys is an explicit alternative that shifts billing to the provider, and the two modes can be mixed (BYOK).
Payment method Card needed before models work

Payment is required before real model calls: the quickstart's second prerequisite is "Credits - ensure your account has sufficient credits", and an empty balance returns 402 Payment Required (First LLM Call, Buying credits). A card is not the only route - credits can be bought by card, PayPal or bank transfer through Stripe, with AWS Marketplace listed as "coming soon" - and testing without paying is possible via sandbox tokens or by using your own provider keys (Buying credits, Sandbox, BYOK).
Gate before models answer No gate

Every catalogue model is callable as soon as you have a key.

No approval, enablement, waitlist, quota or tier step appears between funding an account and calling any catalog model - the quickstart's only prerequisites are a token and credits, the catalog is publicly readable, and provider-less model names are routed automatically (First LLM Call, GET /v3/models, Provider routing). The only gating found is commercial (credits, and the 10 req/s account limit) and organisational (Advanced-plan RBAC over projects and keys) (Rate limits, Users & Organisation).

Everything you need first: An Eden AI account and API token from app.edenai.run, plus either credits or a free sandbox token; no cloud account, provider key, container or cluster is required (First LLM Call, Sandbox).

Running it in production

The same scale applied to the path the vendor recommends for production traffic. Kept separate from the quickstart because for several products here the two are barely related pieces of work.

Shape of the work Change one base URL

Your existing OpenAI-compatible client keeps working. You change a base URL and a key, and nothing else in your code moves.

What production needs: An API token, funded credits (or BYOK provider keys), and awareness of the account rate limit - 10 requests/second by default, shared across all API keys, raised on request, with the organisation owner's limit capping members (First LLM Call, Rate limits, BYOK). Optional production hardening: send traffic to the EU host for residency, create per-environment custom keys with budgets and expiry, and enable the per-project Log Retention toggle only if you actually want prompts stored (EU AI gateway, Custom API keys, Data retention). Advanced-plan features - projects, RBAC, per-environment spend limits, dedicated support with an SLA - require a sales conversation (Users & Organisation, Eden AI pricing).

Can you run it yourself

Install command published No self-hosting

This runs on the vendor’s infrastructure only.

How it fits your stack

2 of 10

Each row is a thing you might already run. “With a caveat” means it works but not the way the vendor’s marketing implies — read the reason, because that is usually where the surprise lives.

Fits The OpenAI SDK Drop-in: change the base URL and key, nothing else.
No The Vercel AI SDK No AI SDK route documented.
No Cloudflare Workers No Workers guidance published.
No Kubernetes No Kubernetes deployment published.
No Terraform or OpenTofu Nothing published for Terraform.
Fits An existing API gateway This is that gateway — AI traffic becomes a plugin, not a new hop.
No Cloud IAM I already run No identity integration published.
With a caveat LangChain or LlamaIndex LangChain only.
With a caveat MCP servers to govern undefined — governs nothing on your side.
With a caveat Nothing — plain Node or Python Change one base URL, but a payment method is needed first.

Reading this the other way round — pick what you already run and see every product scored against it.

The integration surfaces behind those answers

Vercel AI SDK Not documented

Nothing published. Assume the OpenAI-compatible route and verify it yourself.

No Vercel AI SDK page, provider package or createOpenAI-style example appears in the complete documentation index, whose integrations section lists 12 other clients (docs index (llms.txt)), and no *-vercel-provider repository exists in the GitHub organisation (GitHub repos API). The endpoint's OpenAI compatibility makes the SDK's OpenAI-compatible provider a plausible route, but that is inference, not documentation.
Cloudflare Workers Not documented

No Workers guidance either way. If you are edge-first, verify fetch-only compatibility yourself.

As a runtime or deployment target. Cloudflare appears twice in unrelated roles: as one of the 33 LLM serving providers in the catalog (owned_by: cloudflare) (GET /v3/models), and as a named third-party service provider in the privacy policy alongside Hotjar, Intercom, Stripe, Google and Sentry (Privacy policy). Neither states that Eden AI's gateway runs on Workers.
Kubernetes Not documented

No Kubernetes story published.

No Kubernetes, Helm, operator or container guidance exists, consistent with there being no self-hosted component at all - the documentation index has no deployment section (docs index (llms.txt), Eden AI pricing).
Terraform Not documented

No Terraform surface published. Configuration is API or dashboard work.

No Terraform provider, module or registry reference appears in the documentation index or among the 21 public repositories in the GitHub organisation (docs index (llms.txt), GitHub repos API). Account configuration is done in the dashboard and per request, not declaratively (Provider routing).
Existing API gateway It is the API gateway

This product is the gateway. If you already run it for your other APIs, AI traffic becomes a plugin rather than a new hop.

Eden AI is itself the gateway: "a unified AI gateway... through a single API", hosted at https://api.edenai.run/v3 with routing, fallback and cost tracking built in (AI gateway overview). It is not documented as a plugin to any API-gateway platform (docs index (llms.txt)).
Cloud identity Not documented

No identity integration published. Expect API keys in a secret store.

Authentication is a bearer token issued by Eden AI, and no AWS SigV4, IAM role, GCP service account, Azure managed identity, OIDC or workload-identity option appears on the BYOK, custom-keys or organisation pages (BYOK, Custom API keys, Users & Organisation).
MCP

Left empty because no enum value describes an absence here: MCP is not mentioned anywhere - no hosted MCP server, no MCP gateway, no MCP tool parameters, and no docs-MCP endpoint - across the documentation index, the API reference sections and the security page (docs index (llms.txt), Security). The catalog does expose a per-model supports_computer_use capability flag, which is a model attribute passed through, not an Eden AI MCP feature (GET /v3/models). Checked 2026-09-02.

Python frameworks

LangChain

LangChain is the only documented Python framework: pip install "langchain~=1.2" "langchain-openai~=1.1" "langchain-community~=0.4" "langgraph~=1.0", then ChatOpenAI with base_url="https://api.edenai.run/v3"; the same page covers the TypeScript @langchain/openai path (LangChain integration). LlamaIndex, Haystack, DSPy and CrewAI: n.a. - not in the documentation index, although the GitHub organisation holds forks of llama_index and haystack-core-integrations with no Eden AI provider code described (docs index (llms.txt), GitHub repos API).

First-party client libraries

Python
TypeScript
JavaScript

No Eden AI SDK exists; the documented clients are the official OpenAI SDKs with a swapped base URL - openai for Python and openai for TypeScript/JavaScript - plus plain requests/fetch/cURL examples (OpenAI Python SDK, OpenAI TS/JS SDK, First LLM Call). LangChain examples add langchain-openai (Python) and @langchain/openai (TypeScript) (LangChain integration).

Agent features: Solid pass-through of agent primitives, discoverable per model. Tool/function calling is listed as a use case for the LLM endpoint and surfaced as catalog capabilities supports_tools, supports_function_calling, supports_parallel_function_calling and supports_tool_choice, with tool_calls appearing as a streaming finish reason (LLMs vs Expert Models, GET /v3/models, Streaming). Structured output supports both response_format: {"type": "json_object"} and json_schema with strict: true (Structured output). Web search is web_search_options.search_context_size (low/medium/high, default medium), gated by capabilities.supports_web_search, with a documented OpenAI caveat: on /v3/chat/completions it works only on dedicated search variants such as openai/gpt-4o-search-preview, otherwise use the Responses API (Web search). Extended reasoning is Claude-only via thinking.budget_tokens (minimum 1024, and top_p is ignored when enabled) (Chat completions). Statefulness comes from the Responses API with store and previous_response_id (Responses). No MCP and no built-in tool-execution loop are documented (docs index (llms.txt)).

Onboarding is unusually low-friction for a hosted gateway: the model catalog is public and unauthenticated (GET https://api.edenai.run/v3/models returned 1,010 endpoints with per-endpoint pricing, capabilities and regions on 2026-09-02, and GET /v3/embeddings/models likewise), so you can price and capability-check before signing up (GET /v3/models, Embeddings). Sandbox tokens let you wire up the integration for free before funding an account (Sandbox). Two friction points to know about: the expert-model endpoint uses a different model-string grammar (feature/subfeature/provider[/model]) and a different response envelope from the LLM endpoint (LLMs vs Expert Models), and pre-2026 accounts are still on the old app until end of 2026 (AI gateway overview).

Ecosystem coverage is client-side rather than infrastructure-side. Documented integrations: Claude Code, Cline, Codex CLI, Continue.dev, LangChain, LibreChat, n8n (official community node), Open WebUI, OpenAI Python and TS/JS SDKs, OpenClaw and OpenCode (docs index (llms.txt)). The GitHub organisation backs some of these with first-party code - n8n-nodes-edenai, edenai-skill (a Claude Code skill), edenai-openclaw-plugin, pi-edenai and a cookbook of notebooks - alongside forks of third-party projects (GitHub repos API, 2026-09-02). Absent from both: any Vercel AI SDK provider, Terraform, Kubernetes/Helm or self-hosting path (docs index (llms.txt)).

Silence in the docs: 6 of the integration questions on this card have no published answer either way. That is recorded as undocumented, not as a no — but it does mean you would be verifying it yourself.

Actively developed and independent. V3 is the current platform with the pre-2026 app "supported until the end of 2026", so legacy users have a dated migration ([AI gateway overview](https://www.edenai.co/docs/v3/overview/ai-gateway), page dateModified 2026-05-07). The company is French, "founded and headquartered in Lyon", with "25+" employees and "$4.5M raised" ([About Eden AI](https://www.edenai.co/about)); the latest round is a EUR 3M seed led by Galion.exe with 50 Partners and angels including Datadog's Olivier Pomel and Docker's Sebastien Pahl ([Eden AI seed announcement](https://www.edenai.co/post/eden-ai-raises-eu3m-to-bridge-the-gap-between-ai-models-and-business-needs)), which independent coverage dates to late 2024 and reads as early-stage ([Founderland analysis, 2026-04-27](https://www.founderland.ai/articles/eden-ai-launches-european-ai-gateway-to-challenge-openrouter-mogx7fzy)). Docs shipping is visible: the public `edenai/docs` repository was last pushed 2026-09-02 ([GitHub repos API](https://api.github.com/orgs/edenai/repos)). Two caveats for buyers - marketing claims of 99.99% uptime and 200k+ developers are unaudited ([Founderland analysis, 2026-04-27](https://www.founderland.ai/articles/eden-ai-launches-european-ai-gateway-to-challenge-openrouter-mogx7fzy)), and the vendor's own pages disagree on rate limits and model counts ([Plans & Pricing](https://www.edenai.co/docs/v3/overview/plans-prices), [Rate limits](https://www.edenai.co/docs/v3/overview/rate-limits)).

Source

What it does well

No token markup: provider list prices pass through, and 115 of 1,010 catalog endpoints are priced below list via a `discount` field
Multi-modality in one API: 36 expert-model subfeatures (OCR, document parsing, STT/TTS, translation, vision, video generation) alongside 488 routable LLM names
Public, unauthenticated model catalog with per-endpoint pricing, capabilities and region tags
EU-first data path: EU-hosted infrastructure, a dedicated EU endpoint that refuses out-of-EU routing, and per-request region pinning
Zero data retention by default, no training on customer data, with content logging as an opt-in per-project toggle
Per-key budgets, expiry and free sandbox tokens make cost and access control concrete
Ordered per-request fallbacks and cost/latency routing preferences, with failed attempts costing nothing

Where it falls short

Not LLM-routing-first: half the product is non-LLM expert models on a separate, non-OpenAI-compatible endpoint that does not stream
5.5% fee is charged on credit purchases at checkout, so small top-ups pay the fee before a single token is spent
Guardrails are asserted on the marketing security page but have no documented feature, parameter or example anywhere in the docs
No timeout, retry or circuit-breaker controls at all; reliability tuning is limited to a fallback list capped at 3 entries
Rate limits contradict across vendor pages (7, 10 and 15 req/s), and model counts range from 300+ to 500+ depending on the page
No self-hosted or hybrid deployment, no Terraform or Kubernetes path, and no config-as-code surface
Projects, RBAC and per-environment spend limits are Advanced-plan only, on a custom quote with no published price
SOC 2 and ISO 27001 are asserted without a report, portal or audit scope; the subprocessor list is available only on request
No SLA figure, latency claim or throughput claim is published; uptime and developer-count claims are unaudited

Choose it when

Teams that need OCR, speech, translation or document parsing alongside LLM calls under one contract, one prepaid balance and EU data residency, and who are happy to build routing policy per request.

Look elsewhere when

You want an LLM-routing-first gateway with configurable timeouts, retries, enforced guardrails, per-key rate limits or a self-hosted data plane - none of those are documented, and account-level features like projects and RBAC sit behind the quote-only Advanced plan.

Managed marketplace: One account and one key gets you hundreds of models from dozens of providers. Fastest way to start, widest catalog, least control over the data path.

How hard is it to leave?

Derived from six published facts, not from an opinion. The weights are fixed and the same for every product — see the arithmetic.

64 /100 Some work to leave

Portability score breakdown for Eden AI
What helps you leave	Points	Source
Works with standard OpenAI code Switching away is a base-URL change rather than a rewrite of every call site.	22 /22	vendor page
No vendor-specific SDK required A proprietary client library spreads through your codebase and has to be torn out again.	10 /10	—
Can use your own provider accounts Your keys and billing relationship stay yours, so removing the gateway does not cut off model access.	20 /20	vendor page
Can be self-hosted You can run it yourself instead of accepting a pricing or policy change.	0 /20	vendor page
Configuration lives in version control Routing and budget rules are a file you keep, not dashboard state you would have to rebuild.	0 /16	—
Your request history can be exported You leave with your own logs instead of abandoning them.	12 /12	—

Read the fine print: Programmatic export exists for usage and cost data (`/v2/cost_management/`) and uploaded files can be listed and deleted, but there is no documented bulk export of logs, prompts or configuration, and no import/export format for routing settings ([Monitoring](https://www.edenai.co/docs/v3/general/monitoring), [File upload](https://www.edenai.co/docs/v3/llms/file-upload), [docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)).

All six inputs are published, so this is scored against the full 100. This measures technical switching cost only. It does not price the engineering time to re-test prompts against a different routing stack.

Eden AI models & pricing

Browse every imported listing from this provider, with published token rates and a link to compare other providers for the same model. This is provider-reported coverage; an absent listing does not mean unsupported.

Loading model listings…

Official model coverage source ↗ · Model source coverage and limitations

Full specification

Every field we track. Blank fields say "Not published" rather than "No" — we do not infer an absence from silence. Switch to Technical in the header for the precise field names and the low-level details.

Overview

Interpret these fields: Self-hosted vs managed LLM gateways · Who still owns your LLM gateway?

What kind of product Category: Managed marketplace Verified Sep 2, 2026; Marketplaces resell many providers behind one key. Gateways add governance on top. Open-source projects you run yourself. Cloud platforms are hyperscaler surfaces. Inference providers host models on their own hardware.
Who runs it Deployment model: Managed only Verified Sep 2, 2026; Managed means the vendor operates it. Self-host means you run it on your own infrastructure. Both means you can choose.
Licence Licence: Proprietary Verified Sep 2, 2026; Proprietary products cannot be inspected or forked. Open licences such as MIT and Apache-2.0 let you audit, modify, and run the code without permission.
Company Company: Eden AI Verified Sep 2, 2026; The organisation that maintains the product.
Who you would be signing with Vendor status: Independent company Verified Sep 2, 2026; Whether the product is still an independent company, has been acquired, is a large cloud vendor’s product line, is run by a software foundation, or has been put into maintenance mode. Maintenance mode means bug fixes and security patches only — no new features.
Last shipped an update Latest release: Not published Not verified
No dated release artifact could be confirmed. There is no versioned product changelog in the documentation index and the hosted changelog at changelog.edenai.co renders its entries client-side, so no entry date was extractable from either the cleaned text or the raw HTML on 2026-09-02; its visible content still references v2 endpoints such as `https://api.edenai.run/v2/llm/chat` ([Eden AI changelog](https://changelog.edenai.co/)). The freshest dated vendor artifacts observed are the docs-overview `dateModified` of 2026-05-07 ([AI gateway overview](https://www.edenai.co/docs/v3/overview/ai-gateway)) and a 2026-09-02 push to the public docs repository ([GitHub repos API](https://api.github.com/orgs/edenai/repos)).; The date of the most recent release or version tag. A product that has not shipped in a year is a different risk from one that shipped last week, regardless of what its marketing site says.
GitHub stars GitHub stars: Not published Not verified; A rough proxy for community size on open-source projects. Not a quality measure.

Cost

Interpret these fields: How LLM gateway pricing works · LLM gateway spending limits: stop a runaway agent bill?

Markup on model prices Token markup: None Verified Sep 2, 2026; How much the product adds on top of what the underlying model provider charges. Zero means you pay the same per-token price you would pay the model provider directly.
Fee to add funds Credit purchase fee: 5.5% Verified Sep 2, 2026; A percentage charged when you top up your balance, separate from token prices. It is easy to miss because it does not appear on the per-token price list.
Monthly cost per person Seat fee: None Not verified; A recurring per-user platform charge that applies regardless of how much you use the models.
Can use your own provider accounts BYOK supported: Yes Verified Sep 2, 2026; Bring Your Own Key: you keep direct contracts with OpenAI, Anthropic and others, and the gateway only routes traffic. This preserves negotiated rates and committed-spend discounts.
Cost of using your own accounts BYOK terms: No BYOK surcharge is published: the BYOK page states provider billing goes direct and that with BYOK the returned `cost` becomes an estimate rather than a charge; the only fee named anywhere on the pricing pages is the 5.5% credit-purchase fee ([BYOK](https://www.edenai.co/docs/v3/general/byok), [Plans & Pricing](https://www.edenai.co/docs/v3/overview/plans-prices)). Not verified; What the product charges to route traffic through your own provider keys.
Free tier Free tier: No free inference allowance is documented. What is free is the sandbox: sandbox tokens "let you test your integration without real provider calls and at no cost", returning mock data ([Sandbox](https://www.edenai.co/docs/v3/general/sandbox)). Real calls need credits - a depleted balance returns `402 Payment Required` ([Buying credits](https://www.edenai.co/docs/v3/general/buying-credits)). Seats are free: the AI API Gateway plan lists "Unlimited seats" ([Eden AI pricing](https://www.edenai.co/pricing)). Verified Sep 2, 2026; What you can do without paying, useful for evaluation.
Enterprise plan from Enterprise plan from: Not published Not verified; Annual entry price for the enterprise tier, where one is published or credibly reported.
Cost to run it yourself Self-host cost: n.a. - no self-hosted distribution is offered, so there is no self-host cost model. The Advanced plan mentions "private deployments" for specific compliance needs without describing customer-run infrastructure, pricing or artifacts ([Eden AI pricing](https://www.edenai.co/pricing)), and no Docker image, Helm chart, binary or package appears anywhere in the documentation index ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)). Not verified; What self-hosting actually costs once you account for infrastructure and any paid tier.
How the vendor makes money Pricing model: Flat monthly platform fee Verified Sep 2, 2026; The shape of the vendor’s bill: does the routing layer charge a percentage on top of tokens, a flat monthly fee, both, neither (because inference is the product), or nothing at all (open source with no paid tier).
How pricing works, briefly Pricing model detail: The 5.5% is a platform fee on credit purchases, not a per-token markup: "Eden AI charges no markup on provider pricing. You pay exactly what the underlying provider charges, plus a 5.5% Eden AI platform fee applied at checkout" ([Plans & Pricing](https://www.edenai.co/docs/v3/overview/plans-prices)) and the pricing page repeats "We do not mark up provider pricing... The only additional cost is a 5.5% Eden AI platform fee, applied at checkout when purchasing credits" ([Eden AI pricing](https://www.edenai.co/pricing)). So per-request `cost` values returned by the API are provider list prices; the fee is incurred once per credit top-up. Corroborated in the live catalog, where `pricing` equals `list_pricing` for 895 of 1,010 endpoints and is *below* list for the remaining 115 (a `discount` field, e.g. 0.35 on `qwen/qwen3.8-max-0902`), i.e. no endpoint is priced above provider list ([GET /v3/models](https://api.edenai.run/v3/models), 2026-09-02). No subscription: "No subscription, No hidden costs, No API call limit" ([Eden AI pricing](https://www.edenai.co/pricing)). Not verified; A one-paragraph description that covers the caveats a pricing category cannot: introductory rates, per-feature meters, tier gating, and pricing that resets on a specific date.
Minimum commitment Minimum commitment: None stated for the self-serve plan: "No subscription, No hidden costs, No API call limit" ([Eden AI pricing](https://www.edenai.co/pricing)). Credits are prepaid; Advanced/Enterprise accounts can switch to postpaid monthly invoicing with 30-day terms ([Buying credits](https://www.edenai.co/docs/v3/general/buying-credits)). No credit-expiry rule appears in the terms, which do state "Unless otherwise stated, payments are non-refundable" ([Terms of service](https://www.edenai.co/terms)). Not verified; Whether the vendor requires a minimum contract term, a minimum spend, or a provisioned-capacity purchase to get its published rate.
Charges that fire after you go over an allowance Overage terms: No request-volume tiers or overage rates exist - billing is pure pay-per-use against prepaid credits, and each response carries a `cost` field in USD ([AI gateway overview](https://www.edenai.co/docs/v3/overview/ai-gateway)). Running out of credit is a hard stop (`402 Payment Required`), mitigated by auto-refill with a trigger threshold and a monthly refill cap, or by a monthly spend limit on postpaid accounts ([Buying credits](https://www.edenai.co/docs/v3/general/buying-credits)). The Advanced plan's "bulk discounts" and "high rate limits" are unpriced ([Eden AI pricing](https://www.edenai.co/pricing)). Not verified; The line items that scale with usage after an included allowance is exhausted — log storage, extra requests, per-feature meters, data export — which is where cost estimates usually go wrong.
Prompt cache offered Cache mechanism: Both exact and semantic Verified Sep 2, 2026; Whether the gateway offers its own response cache, what kind of match it does (exact request, prefix, semantic), or simply passes provider caching through unchanged.
Discount on cached input Cache-read discount: Not published Not verified; How much cheaper cached tokens are than fresh input, when the vendor publishes a single figure. Bundled-inference clouds usually price this per model instead of as one number.
Premium on cache writes Cache-write premium: Not published Not verified; How much more the first write of a cached prefix costs versus a plain input token. A high write premium and a low hit rate can leave you paying more than you save, so this matters as much as the read discount.
Who captures the cache saving Cache economics: Two distinct mechanisms. (1) Response caching: exact-match on same model plus same input, "enabled by default", toggled per project in the dashboard, and a hit is "returned at no additional cost" - the vendor recommends it for embeddings, moderation, OCR and NER and against chat and image generation ([Caching](https://www.edenai.co/docs/v3/general/caching)). No semantic or similarity cache exists on any page or in the documentation index, which is why the semantic capability is recorded false rather than not-published ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)). (2) Prompt caching: provider-side prefix reuse passed through, with `prompt_cache_key`, `prompt_cache_retention` (`in_memory`, `24h`) and preserved-but-untranslated `cache_control` breakpoints; Eden AI adds default ephemeral boundaries for cache-capable Anthropic and Bedrock Claude models ([Prompt caching](https://www.edenai.co/docs/v3/llms/prompt-caching)). Cached-token prices are per model, exposed as `cache_read_input_token_cost` and `cache_creation_input_token_cost` in the catalog, so no single discount percentage exists ([GET /v3/models](https://api.edenai.run/v3/models)). Not verified; Whether the customer keeps the full saving from caching or the vendor captures part of it — and any conditions attached (write premium, storage fees, best-effort hits).
What you can split spend by Cost attribution: Per provider, per feature/subfeature, per API key and per workflow, via the cost-monitoring API: `GET https://api.edenai.run/v2/cost_management/` takes `begin`/`end` plus a `step` (daily to yearly) and filters on provider, subfeature, token, `workflow_id` and `rag_project_id`, returning total cost, call counts and cost per provider for keys such as `text__chat` and `ocr__ocr`; `/v2/cost_management/credits/` returns the balance ([Monitoring](https://www.edenai.co/docs/v3/general/monitoring)). Every response also carries its own `cost` in USD ([AI gateway overview](https://www.edenai.co/docs/v3/overview/ai-gateway)). Per-user, per-team and per-customer splits are not stated. Not verified; The dimensions the vendor documents for splitting spend — per key, per user, per team, per tag, per customer. Matters if you need to chargeback internally or bill an end customer.
How you get cost data out Cost export: API only (`/v2/cost_management/` and `/v2/cost_management/credits/`), plus the dashboard's cost and performance monitoring ([Monitoring](https://www.edenai.co/docs/v3/general/monitoring), [Eden AI pricing](https://www.edenai.co/pricing)). No CSV download, webhook, S3 or warehouse destination is documented on any fetched page or in the documentation index ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)). Not verified; The mechanisms the vendor publishes for exporting cost and usage data: CSV, an API, webhooks, S3, a data warehouse, or nothing at all. Any per-unit price is included.
Who pays the model bill BYOK mode: Your keys or their credits Verified Sep 2, 2026; Whether you bring your own provider accounts (BYOK), buy inference from this vendor, or can do either. This is the single biggest commercial difference between these products: it decides who holds the contract with the model provider and who carries the spend.

Spend governance in detail

Seven signals matter when a bill starts to hurt: who can spend, how much, on what, and who gets paged when it goes wrong. Everything below is drawn from the vendor’s own pricing and docs pages — how we read these.

Virtual or scoped keys Yes

Keys that carry their own budget and rate-limit policy, so an intern experiment cannot spend against a production budget.

Custom API keys created through `/v2/user/custom_token/`, each with type (`api_token` or `sandbox_api_token`), expiry and its own budget ([Custom API keys](https://www.edenai.co/docs/v3/general/custom-api-keys)).
Budget caps per key Yes

A dollar or token ceiling attached to an individual key. Where enforcement is soft, one over-limit request still completes before the block kicks in.

Per-token `balance` and `active_balance`: the key stops working when its budget reaches $0 ([Custom API keys](https://www.edenai.co/docs/v3/general/custom-api-keys)).
Budget caps per team or workspace Yes — advanced

A ceiling applied at a higher scope than one key — a team, a workspace, a customer, or an entire environment.

Per-project separation with per-environment keys and spend limits is documented as an Advanced-plan feature only ([Users & Organisation](https://www.edenai.co/docs/v3/general/users-organisation)).
Rate limiting as a cost control Not published

Configurable request-per-time-window caps. Platform-set rate limits do not count as spend controls; user-configurable ones do.

Not a customer control: the account-level limit (10 req/s by default, shared across keys) is set by Eden AI and raised on request, with no user-configurable per-key limit documented ([Rate limits](https://www.edenai.co/docs/v3/overview/rate-limits)).
Model allowlists Not published

A policy that constrains which models a key or team can call, keeping expensive frontier models out of the wrong hands.

Not stated as governance. `routing.allowed_providers` restricts which providers a single request may use, which is a routing preference rather than an account policy ([Provider routing](https://www.edenai.co/docs/v3/llms/provider-routing)).
Spend alerts Not published

Alerts fired as spend approaches a threshold. Alerts that only fire after the meter has rolled over are marked as such.

Not stated. Auto-refill has a trigger threshold and a monthly cap, and postpaid accounts can set a monthly spend limit, but no alert or notification feature is documented ([Buying credits](https://www.edenai.co/docs/v3/general/buying-credits)).
Webhook notifications Not published

Programmatic notifications on spend events, so budget breaches can page an on-call or open a ticket.

Not stated; the only webhooks documented deliver async job results ([Webhooks](https://www.edenai.co/docs/v3/expert-models/webhooks)).

Enforcement: Enforced before each request

Source →

Catalog

Interpret these fields: LLM gateway model counts: what “500+” means

Models available Models available: 1,104 Verified Sep 25, 2026; How many models you can call, shown as a range because several vendors publish different totals on different pages. Vendor-reported either way, so counts are not directly comparable — some count every provider variant of the same model separately.
Model providers reachable Upstream providers: 33–68 Verified Sep 2, 2026; How many distinct model providers or labs you can reach, shown as a range where the vendor’s own pages disagree. More providers usually means better redundancy when one has an outage.
Works with standard OpenAI code OpenAI-compatible API: Yes Verified Sep 2, 2026; If yes, you can usually switch to it by changing one base URL, and switch away just as easily. This is the main defence against lock-in.
OpenAI chat endpoint POST /v1/chat/completions: Yes Not verified; The endpoint almost every application ports first. "Not documented" means the vendor never states it, which is different from a documented no.
Anthropic messages endpoint POST /v1/messages: Yes Not verified; Whether Anthropic-shaped calls work without rewriting them. Several products support this only as SDK compatibility or provider passthrough rather than a native endpoint — the detail page says which.
OpenAI Responses endpoint POST /v1/responses: Yes Not verified; The newer stateful OpenAI surface. Support is much thinner across this market than chat completions.
Embeddings endpoint POST /v1/embeddings: Yes Not verified; Whether you can generate vectors through the same gateway, or need a second integration for retrieval workloads.
Image generation endpoint POST /v1/images/generations: Partly Not verified; Whether image models are reachable through the same surface as text.
Audio endpoints POST /v1/audio/*: Partly Not verified; Speech-to-text and text-to-speech. Frequently the first gap in an otherwise complete gateway.
Batch jobs endpoint POST /v1/batches: Not documented Not verified; Asynchronous bulk processing, usually at a discount. Commonly undocumented, and commonly the reason a migration stalls late.
Needs the vendor’s own code library Requires a vendor-specific SDK: No Verified Sep 2, 2026; A proprietary client library spreads through your codebase and has to be torn out again if you leave. “No” is the better answer here, and it means the standard OpenAI client works.
You can export your request history Logs / usage data export: Yes Not verified; Whether you can get your own request logs, traces, or usage records back out — through an API, a bulk export, or a download. Decides whether you leave with your history or abandon it.
Settings can live in version control Declarative config-as-code: No Not verified; Whether routing, fallback, and budget rules can be declared in a file you keep in Git, rather than existing only as settings clicked into a hosted dashboard.
Embeddings Embeddings: Yes Not verified; Text-to-vector models, needed for search and retrieval features.
Image generation Image generation: Yes Not verified; Whether image models are reachable through the same interface.
Speech and audio Speech and audio: Yes Not verified; Text-to-speech or transcription models through the same interface.
Video generation Video generation: Yes Not verified; Whether video models are reachable through the same interface.
Batch processing Batch processing: Not published Not verified; Submitting large jobs for cheaper, slower processing. Often 50% off for work that is not time-sensitive.

Routing & reliability

Interpret these fields: How LLM gateway failover actually works · Does your LLM gateway promise any uptime? · Is routing destroying your prompt cache? · Changing models without breaking production

Uptime it promises in writing Contractual SLA uptime: Not published Not verified; The uptime percentage in a published, contractual service level agreement. A public status page is not an SLA — it reports what happened, it does not promise anything or pay you back when it breaks.
Automatic failover Automatic failover: Yes Verified Sep 2, 2026; When a model provider goes down or rate-limits you, traffic moves to a backup automatically instead of returning errors to your users.
Load balancing Load balancing: Yes Verified Sep 2, 2026; Spreads requests across several providers or keys to raise your effective rate limit.
Rule-based routing Conditional routing: Yes Verified Sep 2, 2026; Send different requests to different models based on rules — for example a cheap model for free users and a strong model for paying ones.
Response caching Response caching: Yes Verified Sep 2, 2026; Reuses the answer when the exact same request comes in again, which cuts both cost and latency.
Similar-question caching Semantic cache: No Verified Sep 2, 2026; Reuses an answer when a new question means roughly the same thing as an earlier one. Saves far more than exact-match caching but can return subtly wrong answers if tuned loosely.
Where you set the timeout Request timeout surface: Not documented Not verified
not_documented for inference: no request-timeout parameter, header or default appears on the chat-completions, fallback, provider-routing or streaming pages ([Chat completions](https://www.edenai.co/docs/v3/llms/chat-completions), [Fallback](https://www.edenai.co/docs/v3/general/fallback), [Streaming](https://www.edenai.co/docs/v3/llms/streaming)). The only timeout published anywhere is on the webhook receiver side - 30 s per delivery attempt with up to 3 retries and exponential backoff capped at 30 s ([Webhooks](https://www.edenai.co/docs/v3/expert-models/webhooks)).; Where a request timeout can be set: per request, in a config file, in the vendor dashboard, or nowhere. Nine of the twenty products documented here do not describe a request timeout at all, so the worst case of a hung upstream call is unknowable from the docs.
Where you set retries Retry policy surface: Not documented Not verified
not_documented: no retry count, backoff strategy or retry-on-status configuration appears on the fallback, provider-routing, chat-completions or rate-limit pages, and there is no retry page in the documentation index ([Fallback](https://www.edenai.co/docs/v3/general/fallback), [Provider routing](https://www.edenai.co/docs/v3/llms/provider-routing), [Chat completions](https://www.edenai.co/docs/v3/llms/chat-completions), [docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)). The documented failure behaviour is fallback to the next candidate, not a retry counter. Default retry count: n.a. Backoff: n.a.; Where retry count and backoff are configured. Worth knowing alongside billing: a retried streaming call can be charged more than once.
Where you set fallbacks Fallback surface: Per request Not verified
ORDERED and per-request: a `fallbacks` array of up to 3 entries (a 4th returns `422`), tried in order, on both `/v3/chat/completions` and Universal AI; entries can pin regions (`model@eu`, `model@us`) and duplicates are not de-duplicated. A second layer is provider-level failover for provider-less model names, switched off with `routing.allow_fallbacks: false`. "Failed attempts cost nothing", and `x-edenai-metadata: enabled` returns per-attempt provider, status and region ([Fallback](https://www.edenai.co/docs/v3/general/fallback), [Create Async Job](https://www.edenai.co/docs/api-reference/universal-ai/create-async-job), observed 2026-09-02).; Where the fallback chain is defined. Almost every product claims fallback; the useful question is whether you can change it from code or only by hand in a dashboard.
Shape of the fallback chain Fallback shape: Ordered list Not verified; An ordered list tries targets in sequence; a weighted split sends a percentage of traffic to each, which is what you need to trial a new model on 5% of requests. Weighted splits are much rarer than the marketing implies.
Upstream health tracking Health checks / circuit breaking: Fixed, cannot change Not verified
not_configurable: health is tracked by Eden AI, not by you. The router "ranks unhealthy providers last" while still keeping them as a last resort, and there is no interval, threshold, ejection window or circuit-breaker setting on the routing, fallback or rate-limit pages ([Provider routing](https://www.edenai.co/docs/v3/llms/provider-routing), [Fallback](https://www.edenai.co/docs/v3/general/fallback)). A public status page is published separately ([Eden AI status](https://app-edenai.instatus.com/)).; Whether the product notices a failing upstream and stops sending traffic to it, and whether you can tune the thresholds. This is what turns a provider outage into a blip rather than a sustained error rate, and only four of the twenty expose it.
Cross-region failover you control Multi-region failover surface: Not published Not verified
Left empty because none of the existing enum values fits: region selection is real but it is per-request and per-endpoint, not a config file, and it is neither undocumented nor non-configurable. You can pin a region on each fallback entry (`model@eu`, `model@us`) ([Fallback](https://www.edenai.co/docs/v3/general/fallback)) or send the whole workload to the EU host `https://api.eu.edenai.run/v3/`, which exposes only EU-compatible providers and returns an error rather than routing outside the EU ([EU AI gateway](https://www.edenai.co/eu)). 275 of 1,010 catalog endpoints carry an `eu` region tag, 378 `us`, 323 `global`, 36 `ap`, 1 `ch` ([GET /v3/models](https://api.edenai.run/v3/models), 2026-09-02).; Whether you can define what happens when a region degrades. A vendor running many regions is not the same as a vendor letting you configure failover between them; only three document a user-controlled mechanism.
Where you set load balancing Load balancing surface: Per request Not verified
Weights are NOT user-settable. For a provider-less model name Eden AI spreads traffic across healthy endpoints with internal weighting and optimises for cost by default; your per-request controls are `routing.sort` (`cost`, `speed`, `latency`, `exact`, also usable as a `model:latency` suffix), `routing.allowed_providers`, `routing.allow_fallbacks` and sticky routing (on by default, to preserve provider prompt caches) ([Provider routing](https://www.edenai.co/docs/v3/llms/provider-routing)). Smart routing goes further: `model: "@edenai"` lets Eden AI pick the model, optionally constrained by `router_candidates` ([Smart routing](https://www.edenai.co/docs/v3/llms/smart-routing)).; Where traffic distribution across upstreams or keys is configured.

Operations

Interpret these fields: LLM gateway observability: traces, logs and export · Running coding agents through an LLM gateway · Changing models without breaking production

Usage dashboards and logs Observability: Yes Verified Sep 2, 2026; Built-in visibility into what was sent, what came back, what it cost, and how long it took.
Spending limits Budget controls: Yes Verified Sep 2, 2026; Hard caps that stop spend before it becomes a surprise invoice. The single most valuable control for a small team.
Rate limits Rate limits: Not published Not verified; Caps on request volume per key or per user, useful for protecting against abuse and runaway loops.
Separate keys per team or app Virtual keys: Yes Verified Sep 2, 2026; Issue scoped keys with their own budgets and permissions so you can attribute cost and revoke access without rotating everything.
Prompt versioning Prompt management: Not published Not verified; Store and version prompts outside your code so they can be changed without a deploy.
Quality testing Evals: Not published Not verified; Built-in tooling to score model output against test cases, so you can tell whether a model swap made things better or worse.
MCP support MCP support: Not published Not verified; Native support for the Model Context Protocol, the emerging standard for connecting models to external tools.
What gets logged Logged content: Your choice Not verified
configurable: an optional "Log Retention" toggle, set per project in the dashboard, switches on storage of full request and response content for debugging and audit; with it off, only metadata is retained ([Data retention](https://www.edenai.co/docs/v3/data-governance/data-retention), observed 2026-09-02). No retention period is published for the opt-in mode.; Whether full prompts and responses are stored, only metadata, or nothing. Full-body logging is the most useful debugging feature here and the one most likely to need a conversation with your compliance team.
You can turn logging off Body-logging opt-out: Yes Not verified
Yes - and the default is already opted out. Content storage is opt-in per project via the Log Retention toggle, so leaving it off (or switching it off) keeps prompts and outputs unstored; GDPR access, deletion and portability requests go to support@edenai.co ([Data retention](https://www.edenai.co/docs/v3/data-governance/data-retention)).; Whether prompt and response bodies can be suppressed while still keeping usage metrics. Nineteen of the twenty document a way to do this; the mechanisms range from a per-request header to an organisation-wide setting.
Traces you can take elsewhere Distributed tracing: Not documented Not verified
not_documented: no OpenTelemetry, OTLP, trace-id or span vocabulary appears on any fetched page or in the documentation index ([Monitoring](https://www.edenai.co/docs/v3/general/monitoring), [docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)). The nearest facility is per-request routing diagnostics - `x-edenai-metadata: enabled` returns each fallback attempt's provider, status and region - which is debugging metadata, not distributed tracing ([Fallback](https://www.edenai.co/docs/v3/general/fallback)).; Whether the product emits OpenTelemetry, a proprietary format, or nothing. OpenTelemetry means the traces land in the tooling you already run instead of only in the vendor’s dashboard.
Where telemetry can go Export destinations: Not published Not verified
n.a. - no OTLP collector, Datadog/Grafana integration, Kafka or S3 sink, or CSV download is documented; usage and cost data leave only through the cost-monitoring REST endpoints ([Monitoring](https://www.edenai.co/docs/v3/general/monitoring), [docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)).; Documented sinks for logs and metrics. This is a good proxy for how replaceable the vendor’s own dashboard is: around twenty destinations means you never have to depend on it, while a CSV download means you do.
Can record user feedback Feedback capture API: No Not verified
n.a. - no feedback, rating, score or annotation endpoint appears on the monitoring page, in the API reference sections listed in the documentation index (chat, responses, embeddings, moderations, universal-ai, files, info, models, cost-monitoring, user-management) or on the pricing page ([Monitoring](https://www.edenai.co/docs/v3/general/monitoring), [docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)).; Whether there is an API to attach a rating or score to a logged request, which is what lets production traffic feed quality work later.
Scores live traffic Online eval hooks: No Not verified
n.a. as a programmatic eval or scoring hook. The only comparison tooling named is a dashboard feature, "Compare models & performances", listed on the self-serve plan ([Eden AI pricing](https://www.edenai.co/pricing)); there is no evals page, dataset, scorer or offline-experiment API in the documentation index ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)).; Whether automated scorers can run against real production requests, rather than only against a test set you assemble yourself.

Performance

Delay it adds Proxy overhead: Not published Not verified; Extra time the product itself adds to each request, on top of however long the model takes. Usually irrelevant next to multi-second model latency, but it matters for high-volume or streaming-sensitive workloads.
Requests per second ceiling Throughput: Not published Not verified; Published sustained request rate before the product becomes the bottleneck. Only relevant at genuinely high volume.
What the request path runs on Architecture class: Undisclosed vendor service Verified Sep 2, 2026
vendor_saas: closed hosted service with no published source or runtime statement. The one architectural clue in the docs is the Anthropic endpoint, described as "native pass-through via litellm" ([Create Anthropic Message](https://www.edenai.co/docs/api-reference/anthropic-messages/create-anthropic-message), [docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)), and the catalog's capability/pricing field names (`supports_prompt_caching`, `cache_read_input_token_cost`) match LiteLLM's model-metadata vocabulary ([GET /v3/models](https://api.edenai.run/v3/models)). No language, runtime or edge-network claim is made on the fetched pages.; The comparable way to talk about latency here. An edge worker, a compiled Go or Rust binary, and a Python proxy have different overhead floors no matter which figures each vendor publishes. Products that never disclose their runtime are recorded as undisclosed rather than assumed.
You can run the request path yourself Self-hostable data plane: Not documented Verified Sep 2, 2026
No Docker image, Helm chart, binary, npm/PyPI package or Terraform artifact for a customer-run data plane appears on any fetched page or in the complete documentation index, which lists integrations (Claude Code, Cline, Codex CLI, Continue.dev, LangChain, LibreChat, n8n, Open WebUI, OpenAI SDKs, OpenClaw, OpenCode) but no deployment section ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt), [Eden AI pricing](https://www.edenai.co/pricing)). The public GitHub organisation's 21 repositories are a docs site, a Claude Code skill, an n8n community node, an OpenClaw plugin, a cookbook and forks of third-party projects - no gateway server ([GitHub org API](https://api.github.com/orgs/edenai), [GitHub repos API](https://api.github.com/orgs/edenai/repos), both queried 2026-09-02).; Whether the component that actually carries your prompts can run on your own infrastructure. Distinct from a vendor offering a self-hosted control plane while still proxying traffic through their network.
Streaming responses Streaming support: Partly Verified Sep 2, 2026
partial by endpoint. The LLM endpoint streams over SSE with `stream: true`, `data:` frames and a `[DONE]` marker, and `stream_options.include_usage: true` adds normalized cache usage and `cost` to the final usage event ([Streaming](https://www.edenai.co/docs/v3/llms/streaming), [Prompt caching](https://www.edenai.co/docs/v3/llms/prompt-caching)). The Universal AI expert-model endpoint is explicitly listed as **not** streaming, so OCR, speech and translation results arrive whole or as async jobs ([LLMs vs Expert Models](https://www.edenai.co/docs/v3/overview/llms-vs-expert-models)). Documented caveat on usage fields: "Fields that the provider does not report can be absent or zero" ([Prompt caching](https://www.edenai.co/docs/v3/llms/prompt-caching)).; Whether token-by-token streaming is documented. The caveats matter more than the yes: some products cannot cancel a stream without still being billed, and several timeout and fallback mechanisms stop applying once the first token has been sent.

Security & compliance

Interpret these fields: LLM gateway compliance: SOC 2, HIPAA and evidence · How LLM gateway guardrails fail · Which LLM gateways store your prompts? · Do you need an MCP gateway as well?

Does your prompt reach their servers Prompt transits vendor: Yes Verified Sep 2, 2026
Always: Eden AI is hosted-only, so every request passes through `api.edenai.run` (or `api.eu.edenai.run`) before reaching the provider, including in BYOK mode where only the billing relationship moves ([AI gateway overview](https://www.edenai.co/docs/v3/overview/ai-gateway), [BYOK](https://www.edenai.co/docs/v3/general/byok)). Eden AI's own infrastructure "is hosted in secure European data centers", while the provider that ultimately serves the request may be elsewhere - provider data-centre locations vary per provider and model and must be checked in the catalog ([Servers location](https://www.edenai.co/docs/v3/data-governance/servers-location)).; Whether the text you send passes through this company’s own infrastructure. If it does, every other promise on this page is a policy commitment rather than a physical impossibility. Self-hosted products can answer no outright.
What they keep if you change nothing Logging default: Metadata only, not content Verified Sep 2, 2026
Content logging is off by default: "By default, Eden AI does not store the content of your requests or responses", and what is kept is metadata - timestamps, token counts and costs - for billing and operations ([Data retention](https://www.edenai.co/docs/v3/data-governance/data-retention)). The terms repeat it: "prompts and outputs are not stored", "Customer Data is processed only as needed to route requests" ([Terms of service](https://www.edenai.co/terms)).; Defaults matter more than options. A product that stores full prompts and replies unless you find the right header will have stored them by the time you read the docs.
How long they keep it Default content retention (days): Nothing kept by default Verified Sep 2, 2026
Zero for request and response content by default - it is not stored at all ([Data retention](https://www.edenai.co/docs/v3/data-governance/data-retention)). Other clocks do exist: async job results are "retained for 7 days", and uploaded files default to 7 days with an `expires_at` timestamp after which "expired files cannot be recovered" ([Data retention](https://www.edenai.co/docs/v3/data-governance/data-retention), [File upload](https://www.edenai.co/docs/v3/llms/file-upload)). Retention periods for the metadata Eden AI does keep, and for the opt-in Log Retention mode, are not published.; Default retention for request content, in days. Zero means nothing is kept. Read the note: several products keep nothing as a rule but make timed exceptions for abuse review or specific models.
Could they train on your prompts Training on customer data: No Verified Sep 2, 2026
Explicitly no, in three places: "Customer Data is not used to train models" and "Customer Data is not sold" ([Terms of service](https://www.edenai.co/terms)), "we do not use your data to train models" ([Privacy policy](https://www.edenai.co/privacy)), and the same statement on the security page ([Security](https://www.edenai.co/security)).; Whether the vendor may use your prompts and outputs to train models. “Not published” means we could not find any position, which is not the same as a no — ask for it in writing.
Where it runs, and what you can pin Region and residency control: EU-first with per-request granularity. Eden AI's infrastructure sits in European data centres ([Servers location](https://www.edenai.co/docs/v3/data-governance/servers-location)); a separate EU endpoint `https://api.eu.edenai.run/v3/` enforces EU residency, exposes only EU-compatible providers and errors rather than routing outside the EU ([EU AI gateway](https://www.edenai.co/eu)); and individual fallback entries can be pinned with `model@eu` or `model@us` ([Fallback](https://www.edenai.co/docs/v3/general/fallback)). Counted in the catalog on 2026-09-02: 275 endpoints tagged `eu`, 378 `us`, 323 `global`, 36 `ap`, 1 `ch` ([GET /v3/models](https://api.edenai.run/v3/models)). The DPA is the counterweight - transfers outside the EEA can occur depending on the providers and features chosen, covered by standard contractual clauses ([Data Processing Agreement](https://www.edenai.co/dpa)). Not verified; Which regions are offered and whether you can force processing to stay in one. A global endpoint that silently picks a region is a different compliance story from an endpoint you pin yourself.
Where safety filters run Guardrail execution location: In the vendor’s cloud Verified Sep 2, 2026
Structurally vendor-side, because Eden AI runs only as hosted SaaS - anything it inspects, it inspects on its own infrastructure ([AI gateway overview](https://www.edenai.co/docs/v3/overview/ai-gateway)). But treat the capability itself as unproven: the only guardrail statement found is a marketing FAQ answer, "Does Eden AI provide guardrails? Yes... including input protection, output moderation, policy-based routing, and fallback rules" ([Security](https://www.edenai.co/security)), and there is no guardrails page, parameter or example anywhere in the documentation index ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)).; A filter that strips personal data only helps if it runs before the data leaves your boundary. If guardrails execute in the vendor’s cloud, the vendor has already received whatever you wanted redacted.
Who else touches the data Subprocessor list: Not published Not verified; The published list of third parties the vendor passes your data to. No list means you cannot know the full chain, which most data-protection agreements require you to.
SOC 2 audited SOC 2 audited: Yes Verified Sep 2, 2026; An independent audit of security controls. Enterprise buyers and their procurement teams routinely require it.
Will sign a HIPAA agreement HIPAA BAA: Not published Not verified; Required before you may send protected health information through the service. Without a signed BAA, healthcare data is off limits.
GDPR commitments GDPR commitments: Yes Verified Sep 2, 2026; Published data processing terms for handling personal data of people in the EU and UK.
Can keep data in the EU EU data residency: Yes Verified Sep 2, 2026; Requests can be processed inside the EU rather than routed to US infrastructure. Often the deciding constraint for European customers.
Does not retain your data Zero data retention: Yes Verified Sep 2, 2026; Prompts and responses are not stored after the request completes. Sometimes a paid add-on rather than the default.
Strips personal data PII redaction: Not published Not verified; Detects and removes identifiers such as names, emails, and card numbers before the request reaches the model provider.
Content guardrails Content guardrails: Yes Not verified; Policy checks on inputs and outputs — blocking unsafe content, enforcing formats, or catching prompt-injection attempts.
Runs fully disconnected Air-gapped deployment: Not published Not verified; Can be deployed in a network with no internet access, which some regulated and defence environments require.
Blocks personal data in prompts PII / DLP enforcement: Not documented Not verified
not_documented as a gateway guardrail. No PII detection, masking or redaction parameter appears on the chat-completions, routing, fallback or security pages, or in the documentation index ([Chat completions](https://www.edenai.co/docs/v3/llms/chat-completions), [Security](https://www.edenai.co/security), [docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)). The closest features are self-service expert models you call yourself - `image/anonymization` (api4ai) and named-entity recognition - not an inline filter on LLM traffic ([GET /v3/info](https://api.edenai.run/v3/info)).; Whether personal data detection sits on the request path and can stop the call, merely inspects and forwards it, or is not documented. A control that only reports is a logging feature, not a policy control.
Blocks prompt injection Injection / jailbreak enforcement: Not documented Not verified
not_documented: no prompt-injection or jailbreak detection appears on any fetched page, and the marketing phrase "input protection" on the security page is not backed by a documented feature ([Security](https://www.edenai.co/security), [docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)).; Whether injection and jailbreak detection can stop a request. Most products offering this call a partner classifier rather than shipping their own.
Blocks harmful content Toxicity / moderation enforcement: Not documented Not verified
not_documented as an enforcing guardrail. Moderation exists as an API you call yourself: OpenAI-compatible `POST /v3/moderations` (default `openai/omni-moderation-latest`, returning `categories`, `category_scores` and `category_applied_input_types`, accepting text and image inputs) and expert-model `text/moderation` from google, microsoft and openai ([Create Moderation](https://www.edenai.co/docs/api-reference/moderations/create-moderation), [GET /v3/info](https://api.edenai.run/v3/info)). Neither is described as intercepting or blocking chat-completions traffic ([Chat completions](https://www.edenai.co/docs/v3/llms/chat-completions)).; Whether hate, violence, sexual and self-harm categories are checked inline and can stop a request, in either direction.
Your own policy rules Custom policy hooks: Not documented Not verified
not_documented: no regex, wordlist, JSON-schema-validation or webhook-classifier guardrail hook is published on any fetched page or in the documentation index ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)).; Whether you can add your own rule — a regex, a webhook, or your own classifier — rather than choosing from the vendor library.
Where guardrails run Guardrail execution location: Not documented Not verified; Whether guardrail evaluation happens inside your infrastructure or on the vendor’s servers. This decides whether the prompt you are trying to protect leaves your network in order to be checked.
If the guardrail itself fails Guardrail failure mode: Not documented Not verified
not_documented: with no guardrail feature specified, there is no fail-open or fail-closed statement to record. The only stated failure semantics concern routing - failed attempts "cost nothing" and fall through to the next candidate ([Fallback](https://www.edenai.co/docs/v3/general/fallback), [Security](https://www.edenai.co/security)).; What happens when the guardrail service times out or errors: does the request proceed unchecked, or is it blocked? This is the worst-documented field in the entire catalogue — only two vendors state it plainly, which means most teams are running a control whose failure behaviour they cannot know.
Third-party guardrail vendors Guardrail integrations: Not published Not verified; Named external guardrail services the product can call. A long list means the product is a router for policy engines rather than a policy engine itself — which also means another vendor bill and another hop.

Compliance evidence

Graded by how strong the evidence is, not whether the word appears on the vendor’s website. An audited report and a marketing claim are different things, and only one of them will satisfy your own auditor.

SOC 2 Claimed, no evidence published "Eden AI is SOC 2 and ISO 27001 certified" on the security page and "SOC 2 & ISO 27001 certified" on the home page; no type, audit period, report or trust portal is named
ISO 27001 Claimed, no evidence published asserted on the same two pages; no certificate number, scope statement or certification body named
GDPR DPA published public DPA last updated May 2026: customer is controller, Eden AI is processor as a "French company operating under European Union data protection laws", with SCCs for outside-EEA transfers and breach notification without undue delay
HIPAA BAA Not published
FedRAMP Not published
ITAR Not published

Vendor source

Fit & integration

Interpret these fields: How much does an LLM gateway lock you in? · Do you need an MCP gateway as well? · Running coding agents through an LLM gateway · Changing models without breaking production

Work to try it Evaluation work shape: Change one base URL Not verified; The shape of the work on the vendor’s own quickstart, from swapping one base URL through to deploying infrastructure. An ordinal class rather than a duration, because elapsed time depends on accounts and quota we cannot see.
Work to run it Production work shape: Change one base URL Not verified; The same scale applied to the vendor’s recommended production path. For several products this is much heavier than the quickstart, which is exactly why both are recorded.
Steps on the quickstart Numbered quickstart steps: 2 Not verified
The quickstart numbers only its two prerequisites (API token from the dashboard, then credits or a sandbox token); the call itself is an unnumbered code block, so "2" is the page's own count rather than a steps-to-first-call figure ([First LLM Call](https://www.edenai.co/docs/v3/quickstart/first-llm-call)).; A literal count of numbered steps on the vendor’s quickstart, recorded as evidence beside the work shape. Zero means the page publishes no numbered procedure at all. Large counts usually mean interleaved language tracks rather than more work.
Can you self-host it today Self-host install documentation: No self-hosting Not verified; Whether an install command is actually published. Several products advertise self-hosting while publishing no command to start from, which a plain yes/no would hide.
Works with the OpenAI SDK OpenAI SDK drop-in: Yes Not verified
Yes, exactly: "full OpenAI API compatibility... a drop-in replacement", with the OpenAI Python and TypeScript SDKs pointed at `base_url="https://api.edenai.run/v3"` and the Eden AI key in place of the OpenAI key ([Chat completions](https://www.edenai.co/docs/v3/llms/chat-completions), [OpenAI Python SDK](https://www.edenai.co/docs/v3/integrations/openai-sdk-python), [OpenAI TS/JS SDK](https://www.edenai.co/docs/v3/integrations/openai-sdk-typescript)). The one boundary is that the expert-model endpoint is not OpenAI-SDK compatible ([LLMs vs Expert Models](https://www.edenai.co/docs/v3/overview/llms-vs-expert-models)).; Whether an existing OpenAI-compatible client can be pointed at it by changing the base URL and key.
Vercel AI SDK support AI SDK provider package: Not documented Not verified
not_documented: no Vercel AI SDK page, provider package or `createOpenAI`-style example appears in the complete documentation index, whose integrations section lists 12 other clients ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)), and no `*-vercel-provider` repository exists in the GitHub organisation ([GitHub repos API](https://api.github.com/orgs/edenai/repos)). The endpoint's OpenAI compatibility makes the SDK's OpenAI-compatible provider a plausible route, but that is inference, not documentation.; Whether a first-party AI SDK provider package exists, or only a community package, a documented workaround, or the generic OpenAI provider pointed at a custom base URL.
Python framework integrations Documented Python frameworks: LangChain Not verified
LangChain is the only documented Python framework: `pip install "langchain~=1.2" "langchain-openai~=1.1" "langchain-community~=0.4" "langgraph~=1.0"`, then `ChatOpenAI` with `base_url="https://api.edenai.run/v3"`; the same page covers the TypeScript `@langchain/openai` path ([LangChain integration](https://www.edenai.co/docs/v3/integrations/langchain)). LlamaIndex, Haystack, DSPy and CrewAI: n.a. - not in the documentation index, although the GitHub organisation holds forks of `llama_index` and `haystack-core-integrations` with no Eden AI provider code described ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt), [GitHub repos API](https://api.github.com/orgs/edenai/repos)).; LangChain, LangGraph, LlamaIndex and similar orchestration frameworks with a documented integration.
Callable from Cloudflare Workers Cloudflare Workers support: Not documented Not verified
not_documented as a runtime or deployment target. Cloudflare appears twice in unrelated roles: as one of the 33 LLM serving providers in the catalog (`owned_by: cloudflare`) ([GET /v3/models](https://api.edenai.run/v3/models)), and as a named third-party service provider in the privacy policy alongside Hotjar, Intercom, Stripe, Google and Sentry ([Privacy policy](https://www.edenai.co/privacy)). Neither states that Eden AI's gateway runs on Workers.; Whether the docs show calling this product from your own Worker. Deliberately separated from the several products whose own gateway runs on Workers, which is a fact about their infrastructure and not about your edge compatibility.
Kubernetes install Helm chart availability: Not documented Not verified
not_documented: no Kubernetes, Helm, operator or container guidance exists, consistent with there being no self-hosted component at all - the documentation index has no deployment section ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt), [Eden AI pricing](https://www.edenai.co/pricing)).; Whether a named, published Helm chart exists, versus Helm being referenced with no chart named, versus generic cluster documentation that has nothing to do with this product.
Terraform support Terraform provider or modules: Not documented Not verified
not_documented: no Terraform provider, module or registry reference appears in the documentation index or among the 21 public repositories in the GitHub organisation ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt), [GitHub repos API](https://api.github.com/orgs/edenai/repos)). Account configuration is done in the dashboard and per request, not declaratively ([Provider routing](https://www.edenai.co/docs/v3/llms/provider-routing)).; Whether you can declare this in code: an official provider, official modules, resources inside a hyperscaler’s provider, a community provider, or only Terraform code shipped in a repo.
Reuses your cloud identity Cloud IAM reuse: Not documented Not verified
not_documented: authentication is a bearer token issued by Eden AI, and no AWS SigV4, IAM role, GCP service account, Azure managed identity, OIDC or workload-identity option appears on the BYOK, custom-keys or organisation pages ([BYOK](https://www.edenai.co/docs/v3/general/byok), [Custom API keys](https://www.edenai.co/docs/v3/general/custom-api-keys), [Users & Organisation](https://www.edenai.co/docs/v3/general/users-organisation)).; Whether you can authenticate with IAM roles, workload identity or managed identities instead of another long-lived API key. Distinguished from products that only accept static upstream provider credentials.
Fits behind your API gateway API gateway integration: It is the API gateway Not verified
Eden AI is itself the gateway: "a unified AI gateway... through a single API", hosted at `https://api.edenai.run/v3` with routing, fallback and cost tracking built in ([AI gateway overview](https://www.edenai.co/docs/v3/overview/ai-gateway)). It is not documented as a plugin to any API-gateway platform ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)).; Whether AI traffic can go through a gateway you already run, and who documented that — the product’s vendor or the gateway’s.
MCP support MCP surface shape: Not published Not verified
Left empty because no enum value describes an absence here: MCP is not mentioned anywhere - no hosted MCP server, no MCP gateway, no MCP tool parameters, and no docs-MCP endpoint - across the documentation index, the API reference sections and the security page ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt), [Security](https://www.edenai.co/security)). The catalog does expose a per-model `supports_computer_use` capability flag, which is a model attribute passed through, not an Eden AI MCP feature ([GET /v3/models](https://api.edenai.run/v3/models)). Checked 2026-09-02.; Which kind of MCP support this is: a gateway that governs many MCP servers, a hosted MCP server you connect to, MCP tools accepted inside the completion API, or client tooling. These are different products behind one acronym.
Needs your own provider key Upstream provider key required: Optional Not verified
Optional. The first call needs only an Eden AI token plus credits (or a free sandbox token) - no provider account ([First LLM Call](https://www.edenai.co/docs/v3/quickstart/first-llm-call)). Bringing your own provider keys is an explicit alternative that shifts billing to the provider, and the two modes can be mixed ([BYOK](https://www.edenai.co/docs/v3/general/byok)).; Whether an upstream provider account and key must exist before your first call works. A prerequisite rather than a step, and it can differ between the hosted and self-hosted forms of the same product.
Gate before models work Model access gate: No gate Not verified
none: no approval, enablement, waitlist, quota or tier step appears between funding an account and calling any catalog model - the quickstart's only prerequisites are a token and credits, the catalog is publicly readable, and provider-less model names are routed automatically ([First LLM Call](https://www.edenai.co/docs/v3/quickstart/first-llm-call), [GET /v3/models](https://api.edenai.run/v3/models), [Provider routing](https://www.edenai.co/docs/v3/llms/provider-routing)). The only gating found is commercial (credits, and the 10 req/s account limit) and organisational (Advanced-plan RBAC over projects and keys) ([Rate limits](https://www.edenai.co/docs/v3/overview/rate-limits), [Users & Organisation](https://www.edenai.co/docs/v3/general/users-organisation)).; Whether an enablement click, a quota grant, a paid tier or an approval form stands between a valid key and a working model call.
Official client languages First-party client SDK languages: Python, TypeScript, JavaScript Not verified
No Eden AI SDK exists; the documented clients are the official OpenAI SDKs with a swapped base URL - `openai` for Python and `openai` for TypeScript/JavaScript - plus plain `requests`/`fetch`/cURL examples ([OpenAI Python SDK](https://www.edenai.co/docs/v3/integrations/openai-sdk-python), [OpenAI TS/JS SDK](https://www.edenai.co/docs/v3/integrations/openai-sdk-typescript), [First LLM Call](https://www.edenai.co/docs/v3/quickstart/first-llm-call)). LangChain examples add `langchain-openai` (Python) and `@langchain/openai` (TypeScript) ([LangChain integration](https://www.edenai.co/docs/v3/integrations/langchain)).; Languages with a first-party client library. An empty list can still mean the product is usable from any language via an OpenAI-compatible SDK.

Additional charges

These are the fees that do not appear on a per-token price list, and they are where estimates usually go wrong.

Eden AI platform fee on credit purchases 5.5%, applied at checkout when buying credits
Advanced AI Platform plan Custom price (quote): high rate limits, bulk discounts, private deployments, dedicated support & SLA, professional services

How pricing actually works

n.a. - no self-hosted distribution is offered, so there is no self-host cost model. The Advanced plan mentions "private deployments" for specific compliance needs without describing customer-run infrastructure, pricing or artifacts ([Eden AI pricing](https://www.edenai.co/pricing)), and no Docker image, Helm chart, binary or package appears anywhere in the documentation index ([docs index (llms.txt)](https://www.edenai.co/docs/llms.txt)).

Common questions

Answered from the fields above, so these move when the catalog moves. Every figure quoted here appears in the specification with its source.

Does Eden AI charge a markup on model prices?

Eden AI adds no percentage markup to model prices. Buying credit costs 5.5%. Other charges on the page: eden ai platform fee on credit purchases (5.5%, applied at checkout when buying credits) and advanced ai platform plan (Custom price (quote): high rate limits, bulk discounts, private deployments, dedicated support & SLA, professional services).

Can Eden AI be self-hosted?

No. Eden AI is available only as a service the vendor operates; there is no self-hosted build. The licence is Proprietary. Prompts therefore leave your network and reach the vendor, which makes its retention and residency terms the control that matters here rather than deployment.

Is Eden AI SOC 2 audited, and will it sign a HIPAA BAA?

A SOC 2 report is available for Eden AI but it does not publish a HIPAA business associate agreement. It offers a GDPR data processing agreement and EU data residency. Each of these is linked to the vendor's own page in the compliance section below.

Does Eden AI retain your prompts?

Eden AI publishes a zero-data-retention position. Whether prompt and response bodies are logged is configurable. Logging can be turned off. Stated retention is zero days. It states that it does not train on customer data. Retention, logging and training on customer data are three separate questions, and a vendor can answer one of them without answering the others.

Can you use your own provider keys with Eden AI?

Yes. Eden AI can route through your own accounts with the underlying model providers, so inference is billed to you directly. No BYOK surcharge is published: the BYOK page states provider billing goes direct and that with BYOK the returned `cost` becomes an estimate rather than a charge; the only fee named anywhere on the pricing pages is the 5.5% credit-purchase fee ([BYOK](https://www.edenai.co/docs/v3/general/byok), [Plans & Pricing](https://www.edenai.co/docs/v3/overview/plans-prices)).

How many models does Eden AI support?

Eden AI publishes no aggregate total; counting its catalog gives 1,104 models, drawn from 33–68 upstream providers. Count of entries returned by the models API on 2026-09-25. Includes every entry exposed by that endpoint; not a count of unique base models. The figure on this page is dated and carries its source.

Official links

Independent coverage

Third-party analysis, walkthroughs and operator threads. We link criticism as readily as praise. Nothing here is written or published by the vendor, by a competitor listed on this site, or by an SEO content farm — how we vet these.

Written reviews and analysis 4

Eden AI review: full-stack AI platform tested & rated Techpoint Africa · Jun 2025 Hands-on sign-up-to-first-call walkthrough of the dashboard and unified API across text, speech, image and document features, from a reviewer who came to it as a working developer rather than an evaluator of gateways.
Eden AI Launches European AI Gateway to Challenge OpenRouter Founderland · Apr 2026 The most sceptical independent read available: it checks the EU-residency positioning against Eden AI's own DPA (transfers outside the EEA are permitted under SCCs) and flags the 99.99% uptime and 200k-developer claims as unaudited.
EdenAI vs OpenRouter: Which AI Aggregator Fits Your Stack? First AI Movers · Nov 2025 Practitioner comparison drawn from hands-on use and an interview with Eden AI's CEO; useful because it lands on the same scope split this entry records - Eden AI for multi-modal workflows, OpenRouter when the stack is LLM-only - and warns that the wide provider catalog can overwhelm small teams.
OpenRouter vs Eden AI: LLM or Multimodal API? QVeris · Jul 2026 Implementation-check guide that makes the structural cost of a cross-modality universal API explicit: OCR pages, speech timestamps, image dimensions, moderation labels, video jobs and LLM streams cannot share one response schema, so cross-modal fallback is usually unsafe.

What has changed here

Sep 25, 2026 Models available Models available 1066 1104 source ↗
Sep 17, 2026 Models available Models available 1027 1066 source ↗
Sep 14, 2026 Models available Models available 500 1027 source ↗
Sep 2, 2026 catalog entry catalog entry Not published Added to the catalog source ↗

See this in the full changelog Back to top ↑

Read the head-to-head

These pairs have a written verdict, not just a table.

Eden AI vs Velokey Read the verdict → Eden AI vs OpenRouter Read the verdict → Eden AI vs Requesty Read the verdict → Eden AI vs Hugging Face Inference Providers Read the verdict →

Usually weighed against

Eden AI vs Hugging Face Inference Providers Compare → Eden AI vs OpenRouter Compare → Eden AI vs Requesty Compare →

Follow updates about Eden AI