ENTERPILOT · NaDdjg · Sep 10, 2026 · Sep 11, 2026 · Sep 11, 2026 · Sep 11, 2026
diff --git a/.env.template b/.env.template
@@ -106,11 +106,11 @@
 # Allow optional /p/{provider}/v1/... passthrough aliases while keeping /p/{provider}/... canonical (default: true)
 # ALLOW_PASSTHROUGH_V1_ALIAS=true
 
-# Comma-separated list of provider types enabled for /p/{provider}/... passthrough (default: openai,anthropic,openrouter,kilo,zai,sglang,vllm,llamacpp,llmd,deepseek,jev)
+# Comma-separated list of provider types enabled for /p/{provider}/... passthrough (default: openai,anthropic,openrouter,kilo,zai,sglang,vllm,llamacpp,llmd,deepseek,edenai,jev)
 # Cohere and audio.cpp native passthrough are opt-in; add cohere or audiocpp when
 # those routes are needed. audio.cpp's native surface includes model management
 # and server-local file paths, so enable it only for trusted callers.
-# ENABLED_PASSTHROUGH_PROVIDERS=openai,anthropic,cohere,openrouter,kilo,zai,sglang,vllm,llamacpp,llmd,audiocpp,deepseek,hetzner,jev
+# ENABLED_PASSTHROUGH_PROVIDERS=openai,anthropic,cohere,openrouter,kilo,zai,sglang,vllm,llamacpp,llmd,audiocpp,deepseek,hetzner,edenai,jev
 
 # Enable the realtime (speech-to-speech) endpoints (default: true): the /v1/realtime
 # websocket (and /p/{provider}/v1/realtime passthrough upgrade), the WebRTC SDP
@@ -726,6 +726,14 @@
 # Optional configured model list; see CONFIGURED_PROVIDER_MODELS_MODE below
 # KILO_MODELS=anthropic/claude-sonnet-4.5,openai/gpt-5.5
 
+# Eden AI (default base URL: https://api.edenai.run/v3)
+# Multi-provider gateway. Model IDs use provider/model and pass through unchanged.
+# EDENAI_BASE_URL is optional: the default above is used when it is unset.
+# EDENAI_API_KEY=
+# EDENAI_BASE_URL=https://api.edenai.run/v3
+# Optional configured model list; see CONFIGURED_PROVIDER_MODELS_MODE below
+# EDENAI_MODELS=openai/gpt-4,anthropic/claude-sonnet-latest
+
 # Z.ai (default base URL: https://api.z.ai/api/paas/v4)
 # For GLM Coding Plan, use: https://api.z.ai/api/coding/paas/v4
 # ZAI_API_KEY=...

diff --git a/README.md b/README.md
@@ -120,6 +120,7 @@ The official SDKs therefore work unchanged. Configure their base URLs as follows
 - Z.ai
 - Alibaba Cloud Model Studio (Bailian)
 - Kilo AI
+- Eden AI
 - MiniMax
 - Xiaomi MiMo
 - OpenCode Go

diff --git a/cmd/gomodel/docs/docs.go b/cmd/gomodel/docs/docs.go
diff --git a/config/config.example.yaml b/config/config.example.yaml
@@ -572,6 +572,21 @@ providers:
     # base_url defaults to "https://llm.chutes.ai/v1".
     # Set base_url when using a different compatible endpoint.
 
+  edenai:
+    type: edenai
+    api_key: "${EDENAI_API_KEY}"
+    # base_url defaults to "https://api.edenai.run/v3".
+    # Multi-provider gateway: model IDs use provider/model notation and pass
+    # through unchanged. Models, context windows, capabilities, and per-token
+    # pricing are discovered from Eden's own /v3/models, and per-request cost
+    # comes from the exact USD "cost" Eden returns on each response, so no
+    # pricing metadata needs declaring here.
+    # GoModel's /v1/responses is served by translating to Eden's chat
+    # completions; Eden's own /v3/responses is a different API and is never used.
+    # models:
+    #   - id: "openai/gpt-4"
+    #   - id: "anthropic/claude-sonnet-latest"
+
   elevenlabs:
     type: elevenlabs
     api_key: "${ELEVENLABS_API_KEY}"

diff --git a/config/config.go b/config/config.go
@@ -128,6 +128,7 @@ func buildDefaultConfig() *Config {
 				"llamacpp",
 				"llmd",
 				"deepseek",
+				"edenai",
 				"jev",
 			},
 		},

diff --git a/config/config_test.go b/config/config_test.go
@@ -132,7 +132,7 @@ func TestBuildDefaultConfig(t *testing.T) {
 	assert.Equal(t, DefaultStreamStallTimeoutSeconds, cfg.Server.StreamStallTimeout)
 	assert.True(t, cfg.Server.EnablePassthroughRoutes)
 	assert.True(t, cfg.Server.AllowPassthroughV1Alias)
-	assert.Equal(t, []string{"openai", "anthropic", "openrouter", "kilo", "zai", "sglang", "vllm", "llamacpp", "llmd", "deepseek", "jev"}, cfg.Server.EnabledPassthroughProviders)
+	assert.Equal(t, []string{"openai", "anthropic", "openrouter", "kilo", "zai", "sglang", "vllm", "llamacpp", "llmd", "deepseek", "edenai", "jev"}, cfg.Server.EnabledPassthroughProviders)
 	assert.Equal(t, ConfiguredProviderModelsModeFallback, cfg.Models.ConfiguredProviderModelsMode)
 	assert.Nil(t, cfg.Cache.Model.Local)
 	assert.Equal(t, 3600, cfg.Cache.Model.RefreshInterval)

diff --git a/docs/advanced/configuration.mdx b/docs/advanced/configuration.mdx
@@ -434,6 +434,7 @@ Set these to automatically register providers. No YAML configuration required.
 | `DEEPSEEK_API_KEY`   | DeepSeek                                           |
 | `OPENROUTER_API_KEY` | OpenRouter                                         |
 | `KILO_API_KEY`       | Kilo AI Gateway                                    |
+| `EDENAI_API_KEY`     | Eden AI                                            |
 | `ZAI_API_KEY`        | Z.ai                                               |
 | `XAI_API_KEY`        | xAI (Grok)                                         |
 | `GROQ_API_KEY`       | Groq                                               |
@@ -446,7 +447,7 @@ Set these to automatically register providers. No YAML configuration required.
 | `LLAMACPP_BASE_URL`  | llama.cpp llama-server / LM Studio (no API key needed unless started with `--api-key`) |
 | `LLMD_BASE_URL`      | llm-d Router/EPP (no API key needed unless its Gateway requires one) |
 
-Most providers can use a custom base URL via `<PROVIDER>_BASE_URL` (for example `OPENAI_BASE_URL`). Chutes AI defaults to `https://llm.chutes.ai/v1` and can be overridden with `CHUTES_BASE_URL`. DeepSeek defaults to `https://api.deepseek.com`; set `DEEPSEEK_BASE_URL` only for a compatible proxy or alternate DeepSeek endpoint. OpenRouter defaults to `https://openrouter.ai/api/v1` and can be overridden with `OPENROUTER_BASE_URL`. Kilo AI defaults to `https://api.kilo.ai/api/gateway` and can be overridden with `KILO_BASE_URL`. Z.ai defaults to `https://api.z.ai/api/paas/v4`; set `ZAI_BASE_URL=https://api.z.ai/api/coding/paas/v4` for the GLM Coding Plan endpoint. SGLang defaults to `http://localhost:30000/v1` when `SGLANG_API_KEY` is set, but keyless deployments should set `SGLANG_BASE_URL` explicitly to register the provider. vLLM follows the same pattern at `http://localhost:8000/v1`. llama.cpp's `LLAMACPP_BASE_URL` is always required (llama-server's default port collides with GoModel's own 8080, so there is no default); `LLAMACPP_API_KEY` is optional. llm-d has no universal endpoint, so `LLMD_BASE_URL` is always required; `LLMD_API_KEY` is optional. Azure uses `AZURE_BASE_URL` for its deployment base URL and accepts an optional `AZURE_API_VERSION` override; otherwise it defaults to `2024-10-21`. Oracle requires `ORACLE_BASE_URL` because its OpenAI-compatible endpoint is region-specific.
+Most providers can use a custom base URL via `<PROVIDER>_BASE_URL` (for example `OPENAI_BASE_URL`). Chutes AI defaults to `https://llm.chutes.ai/v1` and can be overridden with `CHUTES_BASE_URL`. DeepSeek defaults to `https://api.deepseek.com`; set `DEEPSEEK_BASE_URL` only for a compatible proxy or alternate DeepSeek endpoint. OpenRouter defaults to `https://openrouter.ai/api/v1` and can be overridden with `OPENROUTER_BASE_URL`. Kilo AI defaults to `https://api.kilo.ai/api/gateway` and can be overridden with `KILO_BASE_URL`. Eden AI defaults to `https://api.edenai.run/v3` and can be overridden with `EDENAI_BASE_URL`. Z.ai defaults to `https://api.z.ai/api/paas/v4`; set `ZAI_BASE_URL=https://api.z.ai/api/coding/paas/v4` for the GLM Coding Plan endpoint. SGLang defaults to `http://localhost:30000/v1` when `SGLANG_API_KEY` is set, but keyless deployments should set `SGLANG_BASE_URL` explicitly to register the provider. vLLM follows the same pattern at `http://localhost:8000/v1`. llama.cpp's `LLAMACPP_BASE_URL` is always required (llama-server's default port collides with GoModel's own 8080, so there is no default); `LLAMACPP_API_KEY` is optional. llm-d has no universal endpoint, so `LLMD_BASE_URL` is always required; `LLMD_API_KEY` is optional. Azure uses `AZURE_BASE_URL` for its deployment base URL and accepts an optional `AZURE_API_VERSION` override; otherwise it defaults to `2024-10-21`. Oracle requires `ORACLE_BASE_URL` because its OpenAI-compatible endpoint is region-specific.
 
 Every provider type also accepts a comma-separated configured model list via
 `<PROVIDER>_MODELS`, for example `OPENROUTER_MODELS`, `ORACLE_MODELS`,
@@ -594,6 +595,7 @@ export GROQ_API_KEY="gsk_..."        # Registers "groq" provider
 export CHUTES_API_KEY="cpk_..."      # Registers "chutes" provider
 export OPENROUTER_API_KEY="sk-or-..." # Registers "openrouter" provider
 export KILO_API_KEY="..."            # Registers "kilo" provider
+export EDENAI_API_KEY="..."          # Registers "edenai" provider
 export ZAI_API_KEY="..."             # Registers "zai" provider
 # Optional: export ZAI_BASE_URL="https://api.z.ai/api/coding/paas/v4"
 export AZURE_API_KEY="..."           # Registers "azure" provider when paired with AZURE_BASE_URL

diff --git a/docs/docs.json b/docs/docs.json
@@ -226,6 +226,7 @@
                     "providers/multiple-ollama",
                     "providers/kimicode",
                     "providers/hetzner",
+                    "providers/edenai",
                     {
                         "group": "Cloud Platforms",
                         "icon": "cloud",

diff --git a/docs/features/passthrough-api.mdx b/docs/features/passthrough-api.mdx
@@ -136,7 +136,7 @@ from passthrough requests before forwarding them upstream.
 
 Passthrough is intentionally narrow while the API is in beta.
 
-- `openai`, `anthropic`, `openrouter`, `kilo`, `zai`, `sglang`, `vllm`, `llamacpp`, `llmd`, `deepseek`, and `jev`
+- `openai`, `anthropic`, `openrouter`, `kilo`, `zai`, `sglang`, `vllm`, `llamacpp`, `llmd`, `deepseek`, `edenai`, and `jev`
   are enabled by default.
 - Chutes supports passthrough but requires explicit operator opt-in because
   passthrough can forward provider-native routes that do not identify a model.
@@ -160,7 +160,7 @@ Passthrough routes are enabled by default:
 ```env
 ENABLE_PASSTHROUGH_ROUTES=true
 ALLOW_PASSTHROUGH_V1_ALIAS=true
-ENABLED_PASSTHROUGH_PROVIDERS=openai,anthropic,openrouter,kilo,zai,sglang,vllm,llamacpp,llmd,deepseek,jev
+ENABLED_PASSTHROUGH_PROVIDERS=openai,anthropic,openrouter,kilo,zai,sglang,vllm,llamacpp,llmd,deepseek,edenai,jev
 ```
 
 Set `ENABLED_PASSTHROUGH_PROVIDERS` to the provider types you want to expose.
diff --git a/docs/openapi.json b/docs/openapi.json
diff --git a/docs/providers/edenai.mdx b/docs/providers/edenai.mdx
@@ -0,0 +1,219 @@
+---
+title: "Eden AI"
+description: "Configure Eden AI's OpenAI-compatible multi-provider API in GoModel."
+icon: "leaf"
+keywords: ["Eden AI", "edenai", "multi-provider", "OpenAI-compatible", "provider setup"]
+---
+
+Eden AI is a multi-provider gateway exposing an OpenAI-compatible REST API at
+`https://api.edenai.run/v3`. GoModel routes chat completions, streaming, model
+listing, embeddings, and passthrough through the shared OpenAI adapter, so one
+Eden key reaches models from OpenAI, Anthropic, Google, Mistral, Cohere,
+DeepInfra, and others.
+
+## Configure
+
+Create an API key in the [Eden AI console](https://app.edenai.run/) and set:
+
+```bash
+EDENAI_API_KEY=
+```
+
+`EDENAI_BASE_URL` is optional — the provider defaults to
+`https://api.edenai.run/v3`. Set it only to reach a different Eden-compatible
+endpoint:
+
+```bash
+EDENAI_BASE_URL=https://api.edenai.run/v3
+```
+
+<Warning>
+  Use an `https://` endpoint. GoModel **refuses** any Eden request bound for a
+  cleartext destination rather than sending it, because the request body
+  carries the prompt or the embedding input, not just the API key. A
+  non-loopback `http://` base URL therefore fails with an explicit error
+  instead of transmitting anything.
+
+  Redirects are held to the same standard, and only a redirect that stays on
+  the host you configured is followed. A redirect to `http://`, or to any
+  other host — including a subdomain, which Go would otherwise let carry the
+  `Authorization` header — is refused rather than followed, so an
+  upstream-chosen `Location` cannot move your key or your prompt somewhere you
+  did not configure.
+
+  Cleartext to `localhost` (or `127.0.0.1`) is allowed, since that traffic
+  never reaches a network -- this is what keeps a local Eden-compatible proxy
+  usable. If you front Eden with a proxy on another host, terminate TLS on it
+  and point `EDENAI_BASE_URL` at its `https://` address.
+</Warning>
+
+Or in `config.yaml`:
+
+```yaml
+providers:
+  edenai:
+    type: edenai
+    api_key: "${EDENAI_API_KEY}"
+    # base_url: "https://api.edenai.run/v3"
+```
+
+You can also add the credential from the **Providers** page in the admin
+dashboard instead of using env vars.
+
+## Models
+
+GoModel discovers Eden's catalog from Eden's own `GET /v3/models` on startup
+and on every registry refresh. **There is no built-in model list**: a model
+Eden adds is routable as soon as the catalog refreshes, with no GoModel
+upgrade and no configuration change.
+
+Model IDs use `provider/model` notation and are forwarded unchanged:
+
+```json
+{ "model": "openai/gpt-4", "messages": [{ "role": "user", "content": "Hi" }] }
+```
+
+Because the ID already contains a slash, qualify it with the provider name when
+another configured provider exposes the same raw ID:
+`edenai/anthropic/claude-sonnet-latest`. GoModel strips only the outer
+`edenai/` routing qualifier before forwarding.
+
+Optionally pin a configured subset:
+
+```bash
+EDENAI_MODELS=openai/gpt-4,anthropic/claude-sonnet-latest
+```
+
+### Discovered metadata
+
+Each catalog entry contributes metadata that `GET /v1/models` returns and that
+the router, filters, and cost strategies use:
+
+| Eden field | GoModel metadata |
+| ---------- | ---------------- |
+| `context_length` | context window |
+| `capabilities.supports_*` | capabilities, with the prefix stripped (`reasoning`, `function_calling`, `prompt_caching`, …) |
+| `capabilities.input_modalities` | `vision` / `audio` / `video` capabilities |
+| `capabilities.output_modalities` | modes and categories |
+| `pricing` | per-model pricing (see below) |
+
+Output modalities also decide what GoModel advertises: a model whose only
+output is audio or images is left out of `/v1/models`, because Eden here
+serves chat, embeddings, and passthrough only. In practice Eden's catalog
+publishes text output for every model, including the handful that also return
+images, so nothing is currently filtered.
+
+## Pricing
+
+Eden publishes per-token USD rates per model, and GoModel converts them to its
+per-million-token representation (`input_cost_per_token: 6e-8` → `$0.06 /
+MTok`).
+
+Rates come from Eden's `pricing` block, which is what the account is actually
+charged — the undiscounted `list_pricing` with any account discount already
+applied. Each rate is resolved on its own: a usable account rate always wins,
+and `list_pricing` supplies only the individual rates `pricing` does not carry.
+Eden currently publishes the same rates in both blocks, so in practice
+everything resolves from `pricing`; the per-rate fallback is what keeps a
+partially priced model from losing the rates it is missing, which would
+otherwise bill those token types at $0.
+
+A rate Eden reports as `0` is treated as genuinely free rather than missing,
+and a rate neither block publishes usably is left unset rather than invented.
+
+These rates cover input, output, cache reads, and cache writes. Eden's
+context-length-tiered rates (`input_cost_per_token_above_200k_tokens`), its
+`tiered_pricing` list, and its per-query search fees have no GoModel
+equivalent and are not read.
+
+Eden's reasoning and audio rates (`output_cost_per_reasoning_token`,
+`input_cost_per_audio_token`) are also left unread. GoModel would price those
+token types by subtracting the base rate, which assumes the counts are already
+part of the base totals — and Eden's usage object reports only
+`prompt_tokens`, `completion_tokens`, and `total_tokens`, so there is no
+breakdown to confirm that against. Those token types fall back to the base
+input/output rates instead.
+
+Pricing is read live from Eden — nothing is hard-coded, and Eden models do not
+need to be present in GoModel's central model catalog. This makes
+`EDENAI_MODEL_FILTER_MAX_PRICE_PER_MTOK`, cost-based load balancing, and
+price display work for Eden models.
+
+This discovered pricing is what model metadata shows, and what cost accounting
+falls back to when a response carries no exact charge. The exact per-request
+cost Eden returns takes precedence whenever it is available (see below).
+
+## Request cost
+
+Eden returns the exact USD charge for each request as a top-level `cost`
+member — on chat completions and on embeddings alike — and GoModel
+records that figure as the request's cost instead of recomputing it from token
+counts. Eden reprices its upstreams automatically and applies account-level
+discounts, so its own number is authoritative in a way a rate-card
+reconstruction is not.
+
+This makes Eden spend visible to usage records, budgets, cost dashboards, and
+observability, and the usage entry is labelled with the cost source
+`edenai_cost`. If a response carries no usable cost, GoModel falls back to the
+discovered per-model pricing above.
+
+<Note>
+  The `provider` member Eden returns (`"openai"`, `"deepinfra"`) names the
+  upstream Eden routed to. GoModel reports `edenai` as the executing provider —
+  that is the provider it called — and re-exposes Eden's value on the response
+  as `edenai_upstream_provider` so clients can still see which upstream served
+  the request.
+</Note>
+
+## Responses API
+
+GoModel serves `/v1/responses` for Eden by **translating the request to Eden's
+chat-completions endpoint**.
+
+<Warning>
+  Eden's own `/v3/responses` route is **not** the OpenAI Responses API. It takes
+  Eden-specific inputs (`routing`, `router_candidates`, `fallbacks`) and returns
+  its own response object, so GoModel never forwards to it. Treat `/v1/responses`
+  support here as chat-completion translation, not native compatibility.
+</Warning>
+
+## Eden-specific request fields
+
+Eden accepts extra top-level fields on chat completions — `routing`,
+`fallbacks`, `session_id`, `pre_hooks`, and `post_hooks`. GoModel preserves
+unknown top-level JSON fields on chat requests, so these reach Eden unchanged:
+
+```json
+{
+  "model": "openai/gpt-4",
+  "messages": [{ "role": "user", "content": "Hi" }],
+  "fallbacks": ["anthropic/claude-sonnet-latest"]
+}
+```
+
+## Embeddings
+
+Eden's `/embeddings` route is OpenAI-compatible and uses the same
+`provider/model` IDs:
+
+```json
+{ "model": "openai/text-embedding-3-small", "input": "hello" }
+```
+
+Embeddings responses carry the same Eden extensions as chat completions, so the
+exact `cost` Eden reports is recorded for them too. Eden's LLM catalog
+(`GET /v3/models`) does not list embedding models, so embedding IDs are
+forwarded without discovered metadata; pin them under `models:` if you want
+them advertised on `/v1/models`.
+
+## Passthrough
+
+`edenai` is in the default `ENABLED_PASSTHROUGH_PROVIDERS` allowlist, so
+`/p/edenai/...` routes work without operator opt-in. Passthrough is a generic
+forwarder: it sends any path you give it to Eden unchanged, under the
+gateway's own credential.
+
+## Unsupported surfaces
+
+Files, batches, and audio are not exposed for Eden. Requests to those gateway
+endpoints will not route to this provider.
-Original file line number
+Diff line change
@@ Expand Up @@
     - Z.ai
     - Alibaba Cloud Model Studio (Bailian)
     - Kilo AI
+    - Eden AI
     - MiniMax
     - Xiaomi MiMo
     - OpenCode Go
@@ Expand Down @@