archived Visibility internal Owner erik@uvilo.com Approver _ Created _ Updated _

OpenRouter Spec

Overview

This spec describes how to add OpenRouter as a custom endpoint in LibreChat, configure the first model (z-ai/glm-5-turbo), deploy the changes, and verify the integration works end-to-end.

See: OpenRouter Requirements

Step 1: Edit `librechat.yaml`

The source-of-truth config lives at Forge/Configs/LibreChat_Service/librechat.yaml in the uvilo-os repo.

Add an OpenRouter entry under endpoints.custom. This block should be placed alongside existing endpoint configurations:

endpoints:
  # ... existing endpoints ...
  custom:
    - name: "OpenRouter"
      apiKey: "${OPENROUTER_KEY}"
      baseURL: "https://openrouter.ai/api/v1"
      models:
        default: ["z-ai/glm-5-turbo"]
        fetch: true
      titleConvo: true
      titleModel: "z-ai/glm-5-turbo"
      dropParams: ["stop"]
      modelDisplayLabel: "OpenRouter"

Key configuration notes

Field	Value	Reason
`apiKey`	`${OPENROUTER_KEY}`	Must NOT be `OPENROUTER_API_KEY` — that name overrides the OpenAI endpoint
`baseURL`	`https://openrouter.ai/api/v1`	OpenRouter’s OpenAI-compatible API endpoint
`models.default`	`["z-ai/glm-5-turbo"]`	Pre-selected model; the first to enable
`models.fetch`	`true`	Dynamically fetches all available models from OpenRouter
`titleConvo`	`true`	Auto-generates conversation titles
`titleModel`	`"z-ai/glm-5-turbo"`	Uses GLM-5 Turbo for title generation (fast, cheap)
`dropParams`	`["stop"]`	OpenRouter models use varied stop tokens; sending `stop` can cause errors
`modelDisplayLabel`	`"OpenRouter"`	Display name in the LibreChat UI endpoint selector

Step 2: Commit the config change

Stage the updated config: git add Forge/Configs/LibreChat_Service/librechat.yaml
Commit: git commit -m "feat: add OpenRouter custom endpoint with z-ai/glm-5-turbo"
Push the branch

Step 3: Deploy the config

LibreChat reads librechat.yaml at deploy time only — a restart is NOT sufficient.

Copy the updated config to the deployed location:

cp /workspace/erik/uvilo-os/Forge/Configs/LibreChat_Service/librechat.yaml /workspace/librechat/

Redeploy LibreChat:
```
railway redeploy --service LibreChat
```

Step 4: Verify the integration

Reload the browser and start a new conversation
Check the endpoint selector — “OpenRouter” should appear as a selectable endpoint
Select OpenRouter — z-ai/glm-5-turbo should be the default model
Browse models — the model dropdown should show all available OpenRouter models (fetched dynamically)
Send a test message — confirm a response is received from GLM-5 Turbo
Check the OpenRouter dashboard at openrouter.ai/activity — the request should appear with token usage and cost

Step 5: Troubleshoot (if needed)

Symptom	Likely cause	Fix
OpenRouter not in endpoint list	Config not deployed or YAML syntax error	Check `/workspace/librechat/librechat.yaml` is updated; check deploy logs for parse errors
”Invalid API key” error	`OPENROUTER_KEY` env var not set or wrong value	Verify in Railway dashboard → LibreChat service → Variables
No models in dropdown	`fetch: true` missing or OpenRouter API unreachable	Confirm the config has `fetch: true` under `models`; check deploy logs
Model errors / 403	The `stop` param or unsupported params being sent	Ensure `dropParams: ["stop"]` is set
OpenAI endpoint broken	Used `OPENROUTER_API_KEY` as env var name	Rename to `OPENROUTER_KEY` — the `_API_KEY` suffix conflicts with OpenAI

About GLM-5 Turbo

GLM-5 Turbo (z-ai/glm-5-turbo) is Z.ai’s model optimized for agent-driven workflows:

Speed: 200+ tokens per second throughput
Agent optimization: improved instruction decomposition, tool use, and stability across long execution chains
Pricing: $0.96/M input tokens, $3.20/M output tokens
Reasoning: supports reasoning-enabled inference with step-by-step thinking