setup openclaw to use librechat | Friends of the Crustacean 🦞🤝 | Page 1

tiny fog Apr 13, 2026, 5:35 PM

#

I currently have a librechat server, and would like to use openclaw as a provider. How can I enable this endpoint? And in the end how does it work? In librechat will I be able to use slash commands, see tool calls, etc?

karmic phoenixBOT Apr 13, 2026, 5:35 PM

#

strange hollowBOT Apr 13, 2026, 5:35 PM

#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

karmic phoenixBOT Apr 13, 2026, 5:35 PM

#

lucid quarryBOT Apr 13, 2026, 5:39 PM

#

I'm checking the HTTP adapter behavior in source so I don't guess about tool calls or slash commands.

#

Yep, this should work through OpenClaw’s OpenAI-compatible Gateway surface.

Enable it in OpenClaw

{
  gateway: {
    http: {
      endpoints: {
        chatCompletions: { enabled: true },
        // optional, only if you also want /v1/responses
        responses: { enabled: true },
      },
    },
  },
}

#

Main endpoint:

POST /v1/chat/completions

Also served on the same Gateway port:

GET /v1/models
GET /v1/models/{id}
POST /v1/embeddings
POST /v1/responses

Docs:

Auth
Use normal Gateway auth:

Authorization: Bearer <token-or-password>

#

token from gateway.auth.token / OPENCLAW_GATEWAY_TOKEN
password from gateway.auth.password / OPENCLAW_GATEWAY_PASSWORD

Important caveat: this is effectively an operator-level endpoint, so keep it on private ingress only.

LibreChat side
LibreChat custom endpoints are the right fit. Point it at the OpenClaw /v1 base URL, not the full /chat/completions path.

Example librechat.yaml:

endpoints:
  custom:
    - name: "OpenClaw"
      apiKey: "${OPENCLAW_GATEWAY_TOKEN}"
      baseURL: "http://<gateway-host>:<port>/v1"
      models:

#

        default: ["openclaw/default"]
        fetch: true
      modelDisplayLabel: "OpenClaw"

Notes:

LibreChat custom endpoint docs say it appends /chat/completions by default, so /v1 is the right base.
GET /v1/models should return openclaw, openclaw/default, and openclaw/<agentId>.
In OpenClaw, the OpenAI model field is agent selection, not backend provider model selection.

So use:

openclaw/default, or
openclaw/<agentId>

How it behaves in practice

LibreChat will most likely use GET /v1/models plus POST /v1/chat/completions.

#

OpenClaw executes that as a normal agent run internally.
So the selected agent can still use its own tools, policies, and routing.

But for your specific questions:

Slash commands?
Probably no, not as OpenClaw slash commands. Those are native OpenClaw/chat-surface features, not part of the OpenAI-compatible HTTP API.
See tool calls?
For POST /v1/chat/completions, also probably no. The compat endpoint returns normal assistant text, not OpenClaw’s internal tool trace. So LibreChat should behave more like a standard chat frontend than an OpenClaw-native UI.
What about /v1/responses?
OpenClaw does support it, and that endpoint does support client-side function tools / function_call items. But I did not find evidence that LibreChat custom endpoints use /v1/responses by default. Their custom endpoint docs describe the /chat/completions flow.

So my recommendation is:

Enable gateway.http.endpoints.chatCompletions.enabled: true

#

Point LibreChat custom endpoint to http://<gateway-host>:<port>/v1
Use openclaw/default as the model
Expect working chat through OpenClaw agents, but not native OpenClaw slash-command UX or full internal tool-call visibility

If you want, I can also sketch a complete OpenClaw + LibreChat config example for Docker or bare metal.

tiny fog Apr 13, 2026, 5:45 PM

#

How about librechat params titleModel, if I have openrouter setup as another provider in librechat, can I use that instead of openclaw for titleModel?

lucid quarryBOT Apr 13, 2026, 5:46 PM

#

I'll verify the titleModel / titleEndpoint behavior in LibreChat docs so I can give you the exact answer.

#

Yes, but not with titleModel alone.

If your OpenClaw custom endpoint is the chat endpoint, then:

titleModel by itself uses the same endpoint
so if you want titles generated by OpenRouter instead of OpenClaw, set titleEndpoint to your OpenRouter custom endpoint name, then set titleModel to an OpenRouter model

Example:

endpoints:
  custom:
    - name: "OpenClaw"
      apiKey: "${OPENCLAW_GATEWAY_TOKEN}"
      baseURL: "http://<gateway-host>:<port>/v1"
      models:

#

        default: ["openclaw/default"]
        fetch: true
      titleConvo: true
      titleEndpoint: "OpenRouter"
      titleModel: "openai/gpt-4o-mini"

    - name: "OpenRouter"
      apiKey: "${OPENROUTER_KEY}"
      baseURL: "https://openrouter.ai/api/v1"
      models:
        fetch: true
        default: ["openai/gpt-4o-mini"]

So the rule is:

#

same endpoint for titles → just set titleModel
different provider/endpoint for titles → set both titleEndpoint + titleModel

LibreChat docs mention exactly this “use a different custom endpoint for titles” pattern:
https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/custom_endpoint

If you want, I can write the full LibreChat config block for:

OpenClaw for chats
OpenRouter for titles
and optionally OpenRouter for summaries too

#setup openclaw to use librechat