Skip to content

Latest commit

 

History

History
125 lines (98 loc) · 5.69 KB

OPENCLAW-INTEGRATION.md

File metadata and controls

125 lines (98 loc) · 5.69 KB

OpenClaw Integration

OpenModelStudio can be driven by OpenClaw AI agents so users talk in natural language (e.g. Telegram, Discord, WhatsApp) and the agent creates projects, starts training, and manages experiments via the API. The same REST API (and API key auth) works for any coding agent or CLI—see API for agents & automation.

Overview

  • OpenClaw runs the agent (LLM + tools) and handles channels (Telegram, Discord, etc.).
  • An OpenModelStudio plugin for OpenClaw registers tools that call the OpenModelStudio REST API.
  • The user sees the agent “doing everything” in the chat; the same actions appear in the OpenModelStudio UI (projects, jobs, experiments).

Plugin location

The plugin lives in the repo at:

integrations/openclaw/
├── openclaw.plugin.json   # Manifest + config schema
├── package.json           # npm package + OpenClaw extension entry
├── index.ts               # Tool definitions + OMS API client
└── README.md              # Install, config, tool list, SOUL.md tips

Quick setup

  1. Install OpenClaw (see openclaw.im or docs.openclaw.ai).

  2. Install the plugin (from OpenModelStudio repo root):

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

  3. Get an OpenModelStudio API key or JWT
    Recommended: In the OpenModelStudio UI, go to Settings → API Keys, create a key (e.g. "OpenClaw"), and use that as accessToken. It never expires and end users don’t need to provide a token.
    Alternatively: Get a JWT (default test user):

    curl -s -X POST http://localhost:31001/auth/login \
  -H "Content-Type: application/json" \
  -d '{"email":"test@openmodel.studio","password":"Test1234"}' | jq -r '.access_token'

  4. Configure OpenClaw (e.g. ~/.openclaw/openclaw.json):

    {
  plugins: {
    entries: {
      "openclaw-plugin": {
        enabled: true,
        config: {
          baseUrl: "http://localhost:31001",
          accessToken: "oms_xxxx..."  // or JWT from login
        }
      }
    }
  },
  agents: {
    list: [{
      id: "main",
      tools: { allow: ["openclaw-plugin"] }
    }]
  }
}

    (Entry key openclaw-plugin comes from the package name.)

  5. Restart the OpenClaw Gateway. The agent can then use tools like oms_create_project, oms_start_training, oms_create_experiment, etc.

Tools registered by the plugin

Tool Purpose
oms_list_projects List projects
oms_create_project Create project (name, description?)
oms_list_models List models (optional project_id)
oms_create_model Create model (project_id, name, framework, …)
oms_start_training Start training job (model_id, …)
oms_list_training_jobs List training jobs
oms_get_training_job Get job status
oms_get_job_metrics Get job metrics
oms_create_experiment Create experiment
oms_add_experiment_run Add run (experiment_id, job_id, params, metrics)
oms_list_experiment_runs List runs for an experiment
oms_list_experiments List experiments
oms_launch_workspace Launch JupyterLab workspace
oms_list_datasets List datasets
oms_search Search projects/models/datasets

All tools are optional in OpenClaw; allowlist them (or the plugin id openmodelstudio) under agents.list[].tools.allow.

Agent instructions

In your OpenClaw workspace, add instructions (e.g. in SOUL.md or AGENTS.md) so the agent uses the tools in a sensible order:

  • Create project → create model → start training → create experiment → add run.
  • Use IDs returned by the API (project_id, model_id, job_id, experiment_id) in later tool calls.
  • After each action, confirm what was done and point the user to the OpenModelStudio dashboard.

See integrations/openclaw/README.md for an example SOUL.md snippet.

Architecture

User (Telegram/Discord/…)
    → OpenClaw (agent + channels)
        → Plugin tools (oms_*)
            → HTTP GET/POST to OpenModelStudio API (Bearer API key or JWT)
                → OpenModelStudio (Rust API, DB, K8s, …)

The plugin does not run inside OpenModelStudio; it runs inside the OpenClaw Gateway and only needs the API base URL and a valid API key or JWT.

Security

  • accessToken is sensitive; store it in OpenClaw config (or use Control UI with sensitive hint). Do not commit tokens.
  • Use an API key (Settings → API Keys) so end users never need to provide a token; all tool calls run as the user who created the key. Use a dedicated user or key for the agent if you want to audit or restrict access.
  • OpenModelStudio API enforces auth and scoping (e.g. projects owned by or shared with the user).
  • API keys are stored only as a SHA-256 hash in the database; the raw key is never persisted.

Production: Use HTTPS for the OpenModelStudio API (and OpenClaw gateway if exposed) so tokens and API keys are never sent in the clear. Practice good secret handling: keep API keys and tokens in a secrets manager or secure config, restrict file permissions on openclaw.json, and never commit credentials to version control.

See also