Skip to content

Latest commit

 

History

History
129 lines (85 loc) · 4.01 KB

OPENCLAW-AND-CLAUDE.md

File metadata and controls

129 lines (85 loc) · 4.01 KB

OpenClaw and Claude: Step-by-Step Guide

This guide walks you through controlling OpenModelStudio with OpenClaw (chat bots) and Claude Code (terminal) using the same API and API key.

Prerequisites

  • OpenModelStudio running (API reachable).
  • An API key from OpenModelStudio: log in to the UI → Settings → API Keys → Create (e.g. name: OpenClaw). Copy the key once (oms_...).

Part 1: OpenClaw (Telegram / Discord / Control UI)

OpenClaw is an AI agent that runs 24/7. You talk to it in natural language; it uses tools to call the OpenModelStudio API.

Step 1 — Start OpenModelStudio

cd /path/to/OpenModelStudio
make k8s-deploy

Or API only: make dev-api → API at http://localhost:8080.

Step 2 — Install the OpenClaw plugin

openclaw plugins install -l ./integrations/openclaw
cd ~/.openclaw/extensions/openmodelstudio && npm install

Step 3 — Configure the plugin

Edit ~/.openclaw/openclaw.json. Under plugins.entries["openclaw-plugin"].config set:

"openclaw-plugin": {
  "enabled": true,
  "config": {
    "baseUrl": "http://localhost:31001",
    "accessToken": "oms_your_api_key_here"
  }
}
  • Use http://localhost:8080 if you used make dev-api.
  • Allow the plugin for your agent, e.g. agents.list[].tools.allow: ["openclaw-plugin"].

Step 4 — Start OpenClaw and add LLM auth

openclaw gateway
  • Control UI: http://127.0.0.1:18789
  • Add your Anthropic (or other) API key for the agent: openclaw agents add main or via the UI.

Step 5 — Use it

In Telegram, Discord, or the Control UI, say things like:

  • "List my OpenModelStudio projects."
  • "Create a project called Sales and add a dataset named sales_data."
  • "Start training for model X with dataset Y."

The agent uses the oms_* tools and reports back. No user needs to paste a token.

Part 2: Claude Code (terminal)

Claude Code is the Claude CLI in your terminal. You give it the same base URL and API key; it runs curl or scripts to call the OpenModelStudio API.

Step 1 — Start OpenModelStudio

Same as Part 1 — e.g. make k8s-deploy (API at http://localhost:31001) or make dev-api (http://localhost:8080).

Step 2 — Set environment variables

In the same terminal where you will run claude:

export OMS_BASE_URL="http://localhost:31001"
export OMS_API_KEY="oms_your_api_key_here"

Use http://localhost:8080 if you used make dev-api.

Step 3 — Start Claude Code

claude

Tell Claude: "Use the env vars OMS_BASE_URL and OMS_API_KEY for OpenModelStudio API calls."

Step 4 — Ask Claude to control OpenModelStudio

Examples:

  • "List my OpenModelStudio projects."
  • "Create an OpenModelStudio project named Test."
  • "Create a dataset in project <project_id> named mydata."
  • "Create dummy CSV data and upload it to OpenModelStudio dataset <dataset_id>."
  • "Create a PyTorch model in project <project_id> named MyModel."
  • "Start OpenModelStudio training for model <model_id> with dataset <dataset_id>."
  • "Get status of OpenModelStudio training job <job_id>."
  • "Get metrics for OpenModelStudio training job <job_id>."
  • "Launch an OpenModelStudio workspace for project <project_id>."

Claude will call the API (e.g. with curl or the Python SDK) using your base URL and API key.

Summary

OpenClaw Claude Code
Where Telegram, Discord, Control UI Terminal (claude)
Config ~/.openclaw/openclaw.json → baseUrl + accessToken OMS_BASE_URL + OMS_API_KEY in shell
Auth API key in config (no user-provided token) API key in env
Same API Yes Yes

Both use the same REST API and API key. Production: use HTTPS and good secret handling; see OpenClaw Integration and API for agents.