Integrations

OpenClaw

OpenClaw is an open-source assistant that can plug into custom OpenAI- or Anthropic-compatible providers. Add Pendra as a provider in openclaw.json and reference it by ID (pendra/qwen3.6:27b).

OpenClaw is an independent project — not affiliated with Pendra or Anthropic. The snippets below are based on OpenClaw's current documentation; double-check against docs.openclaw.ai when you set it up.

1. Add Pendra as a custom provider

Edit your openclaw.json config (JSON5 format):

openclaw.json

// openclaw.json (JSON5)
{
  models: {
    mode: "merge",
    providers: {
      "pendra": {
        baseUrl: "https://api.pendra.ai/api/v1",
        apiKey: "${PENDRA_API_KEY}",
        api: "openai-completions",
        models: [
          {
            id: "qwen3.6:27b",
            name: "Qwen 3.6 27B (Pendra)",
            contextWindow: 32768,
            maxTokens: 8192,
          },
        ],
      },
    },
  },
}

What each key does

Key	Purpose
`mode: "merge"`	Adds Pendra alongside OpenClaw's built-in catalogue. Use `"replace"` to make Pendra the only provider.
`baseUrl`	Pendra's OpenAI-compatible API base.
`apiKey`	Your Pendra `pdr_sk_` key. Use `${ENV_VAR}` to pull from an environment variable.
`api`	`"openai-completions"` for the Chat Completions endpoint, or `"anthropic-messages"` to use Pendra's Anthropic surface (see below).
`models`	The list of models you want exposed under this provider — each becomes `pendra/<id>` in OpenClaw.

 2. Set the key and run
       bash 
  
 export PENDRA_API_KEY=pdr_sk_...

openclaw --model pendra/qwen3.6:27b
 
  Anthropic mode
 
Prefer the Anthropic Messages wire format? Add a second provider with
api: "anthropic-messages" pointing at the bare Pendra host —
    OpenClaw will hit /v1/messages.
       openclaw.json (snippet) 
  
 "pendra-claude": {
  baseUrl: "https://api.pendra.ai",
  apiKey: "${PENDRA_API_KEY}",
  api: "anthropic-messages",
  models: [
    { id: "qwen3.6:27b", name: "Qwen 3.6 27B" },
  ],
},
 
  Choosing models
 
Each entry under models needs an id that matches
    a model serving on your Pendra worker pool. List what's available at
/models; pick reasoning-strong models for
    coding tasks (qwen3.6:27b, gpt-oss:120b, larger
    Llama variants).
 Troubleshooting
  "Provider not found": check the key in models.providers matches what you reference in --model pendra/<id>.
 401 Unauthorized: make sure PENDRA_API_KEY (or whichever env var you used) is exported.
 Stale catalogue: OpenClaw caches provider metadata. Restart it after editing openclaw.json.
 
 Related
  Chat completions reference
 Anthropic Messages reference