OpenClaw

What is OpenClaw

OpenClaw is an open-source, self-hosted personal AI assistant and gateway. It connects chat channels such as WhatsApp, Slack, Discord, Telegram, iMessage, and WebChat to AI agents. This setup adds Capriole AI as three custom model providers, one for each API protocol that Capriole AI serves.

Prerequisites

Before you begin, make sure you have:

A Capriole AI account (Sign up here)
A Capriole AI API key (Get your key)

Installation

Install OpenClaw with the official install script:

curl -fsSL https://openclaw.ai/install.sh | bash

Then complete OpenClaw’s own setup:

openclaw onboard

Configuration

Model validation

OpenClaw 2026.4.21 passes a file-read, file-write, and final-answer check through Capriole AI with every model in Supported models, using the configuration below. Add the models block and the agents.defaults model settings below to ~/.openclaw/openclaw.json. Keep the other settings that openclaw onboard wrote.

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "capriole-ai-anthropic/claude-latest"
      },
      "models": {
        "capriole-ai-openai/openai-latest": {},
        "capriole-ai-openai/openai/gpt-6.1-sol": {},
        "capriole-ai-anthropic/claude-latest": {},
        "capriole-ai-anthropic/claude-opus-5-5": {},
        "capriole-ai-compatible/zai/glm-5.3-flash": {},
        "capriole-ai-compatible/moonshot/kimi-k3.1-pro": {},
        "capriole-ai-compatible/moonshot/kimi-k3": {},
        "capriole-ai-compatible/xai/grok-4.7": {},
        "capriole-ai-compatible/google/gemini-3.8-flash": {}
      }
    }
  },
  "models": {
    "mode": "merge",
    "providers": {
      "capriole-ai-openai": {
        "baseUrl": "https://api.caprioletech.com/v1",
        "apiKey": "${CAPRIOLE_AI_API_KEY}",
        "api": "openai-responses",
        "models": [
          {
            "id": "openai-latest",
            "name": "OpenAI latest",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 1050000,
            "maxTokens": 128000,
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
            "api": "openai-responses"
          },
          {
            "id": "openai/gpt-6.1-sol",
            "name": "GPT-6.1 Sol",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 1050000,
            "maxTokens": 128000,
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
            "api": "openai-responses"
          }
        ]
      },
      "capriole-ai-anthropic": {
        "baseUrl": "https://api.caprioletech.com",
        "apiKey": "${CAPRIOLE_AI_API_KEY}",
        "api": "anthropic-messages",
        "request": {
          "auth": {
            "mode": "authorization-bearer",
            "token": "${CAPRIOLE_AI_API_KEY}"
          }
        },
        "models": [
          {
            "id": "claude-latest",
            "name": "Claude latest",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 1000000,
            "maxTokens": 128000,
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
            "api": "anthropic-messages"
          },
          {
            "id": "claude-opus-5-5",
            "name": "Opus 5.5",
            "reasoning": false,
            "input": ["text"],
            "contextWindow": 1000000,
            "maxTokens": 128000,
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
            "api": "anthropic-messages"
          }
        ]
      },
      "capriole-ai-compatible": {
        "baseUrl": "https://api.caprioletech.com/v1",
        "apiKey": "${CAPRIOLE_AI_API_KEY}",
        "api": "openai-completions",
        "models": [
          {
            "id": "zai/glm-5.3-flash",
            "name": "GLM 5.3 Flash",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 1000000,
            "maxTokens": 128000,
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
          },
          {
            "id": "moonshot/kimi-k3.1-pro",
            "name": "Kimi K3.1 Pro",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 1048576,
            "maxTokens": 131072,
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
          },
          {
            "id": "moonshot/kimi-k3",
            "name": "Kimi K3",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 1048576,
            "maxTokens": 131072,
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
          },
          {
            "id": "xai/grok-4.7",
            "name": "Grok 4.7",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 500000,
            "maxTokens": 128000,
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
          },
          {
            "id": "google/gemini-3.8-flash",
            "name": "Gemini 3.8 Flash",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 1048576,
            "maxTokens": 65536,
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
          }
        ]
      }
    }
  }
}

What each setting does:

capriole-ai-openai sends GPT models to the Responses API. capriole-ai-anthropic sends Claude models to Anthropic Messages. capriole-ai-compatible sends Gemini, Grok, GLM, and Kimi models to Chat Completions.
${CAPRIOLE_AI_API_KEY} reads your key from the environment.
request.auth with authorization-bearer makes the Anthropic provider send the key as a Bearer token. Without it, the key goes in an x-api-key header, which Capriole AI does not accept.
The Anthropic baseUrl has no /v1, because OpenClaw adds /v1/messages itself.
reasoning is false for Opus 5.5. When it is true, OpenClaw sends a thinking budget, and Opus 5.5 rejects it with requires adaptive thinking.
contextWindow and maxTokens tell OpenClaw each model’s limits.
agents.defaults.models is the list of models that agents can use. agents.defaults.model.primary is the default.

Run OpenClaw

Export your key in every terminal that starts OpenClaw:

export CAPRIOLE_AI_API_KEY="YOUR_CAPRIOLE_AI_API_KEY"

Start the gateway:

openclaw gateway

Open the terminal UI in another terminal:

openclaw tui

For a one-off local run without the gateway:

openclaw agent --local --session-id capriole-test --message "Summarize the files in my workspace."

Supported models

Model references use the form <provider-id>/<model-id>.

Model	Model reference	Protocol
OpenAI latest	`capriole-ai-openai/openai-latest`	Responses
GPT-6.1 Sol	`capriole-ai-openai/openai/gpt-6.1-sol`	Responses
Claude latest	`capriole-ai-anthropic/claude-latest`	Anthropic Messages
Opus 5.5	`capriole-ai-anthropic/claude-opus-5-5`	Anthropic Messages
Gemini 3.8 Flash	`capriole-ai-compatible/google/gemini-3.8-flash`	Chat Completions
Grok 4.7	`capriole-ai-compatible/xai/grok-4.7`	Chat Completions
GLM 5.3 Flash	`capriole-ai-compatible/zai/glm-5.3-flash`	Chat Completions
Kimi K3.1 Pro	`capriole-ai-compatible/moonshot/kimi-k3.1-pro`	Chat Completions
Kimi K3	`capriole-ai-compatible/moonshot/kimi-k3`	Chat Completions

openai-latest and claude-latest always point to our latest flagship in each family.

Troubleshooting

Why does authentication fail?

Export CAPRIOLE_AI_API_KEY in the terminal that starts openclaw gateway or openclaw agent. OpenClaw prints missing env var when the variable is not set.

Why does Opus 5.5 fail with `requires adaptive thinking`?

Set reasoning to false for claude-opus-5-5.

Why does OpenClaw report that a model is not allowed or missing?

Run openclaw models list. Confirm that the model reference is in agents.defaults.models, and that the model entry has the full object shown above, not only id and name.

Why does a new session start with a setup conversation?

That is OpenClaw’s first-run workspace setup, not a Capriole AI issue. Finish it once in openclaw tui, or run openclaw onboard again.

Why doesn’t the gateway start?

Run openclaw status, openclaw gateway status, openclaw logs --follow, and openclaw doctor.

Notes

To add another Capriole AI model, copy the entry of a model from the same family, change the ID, and add its reference to agents.defaults.models. GET https://api.caprioletech.com/v1/models lists every model ID.
Official references: OpenClaw install, OpenClaw custom providers, OpenClaw model providers, and OpenClaw troubleshooting.

Get started

Endpoints

Integration

What is OpenClaw

Prerequisites

Installation

Configuration

Model validation

Run OpenClaw

Supported models

Troubleshooting

Why does authentication fail?

Why does Opus 5.5 fail with `requires adaptive thinking`?

Why does OpenClaw report that a model is not allowed or missing?

Why does a new session start with a setup conversation?

Why doesn’t the gateway start?

Notes

​What is OpenClaw

​Prerequisites

​Installation

​Configuration

​Model validation

​Run OpenClaw

​Supported models

​Troubleshooting

​Why does authentication fail?

​Why does Opus 5.5 fail with requires adaptive thinking?

​Why does OpenClaw report that a model is not allowed or missing?

​Why does a new session start with a setup conversation?

​Why doesn’t the gateway start?

​Notes

What is OpenClaw

Prerequisites

Installation

Configuration

Model validation

Run OpenClaw

Supported models

Troubleshooting

Why does authentication fail?

Why does Opus 5.5 fail with `requires adaptive thinking`?

Why does OpenClaw report that a model is not allowed or missing?

Why does a new session start with a setup conversation?

Why doesn’t the gateway start?

Notes