Any model. Works with Claude, GPT, and open-source models.
Run anywhere. Webhooks, built-in cron, approvals, CI/CD, Mac, Linux, Windows or Docker.
No SDK required. Define your agent in Markdown.

Quick Start • Example • Deploy • Documentation

# Try it now - no install needed
ANTHROPIC_API_KEY=sk-ant-... npx agentuse@latest run https://agentuse.io/hello.agentuse

Create my-agent.agentuse:

---
model: anthropic:claude-sonnet-4-6
---

Generate a daily motivation quote with a tech fact.
Format as JSON with 'quote' and 'fact' fields.

Run it:

agentuse run my-agent.agentuse

A real-world agent with MCP tools:

---
model: anthropic:claude-sonnet-4-6
mcpServers:
  postgres:
    command: npx
    args: ["-y", "@modelcontextprotocol/server-postgres"]
    requiredEnvVars: [DATABASE_URL]
---

Query the sales table for yesterday's metrics.
Generate an executive summary with trends.

Webhook Server - Trigger agents via HTTP:

agentuse serve
curl -X POST http://localhost:12233/run -d '{"agent": "my-agent"}'

# Serve multiple projects from one process:
agentuse serve -C ./projA -C ./projB
curl -X POST http://localhost:12233/run -d '{"project":"projA","agent":"my-agent"}'

Global config - put serve defaults in ~/.agentuse/config.json:

{
  "serve": {
    "projects": [
      { "path": "~/work/projA" },
      { "id": "b", "path": "~/work/projB" }
    ],
    "default": "projA",
    "port": 12233,
    "host": "127.0.0.1",
    "auth": true,
    "logFile": true
  }
}

CLI flags override config. -C replaces serve.projects; AGENTUSE_CONFIG=/path/to/config.json uses another file. AGENTUSE_API_KEY remains env-only.

Scheduled Agents - Run on a schedule:

---
schedule: "0 9 * * *"
---

Approval Gates - Pause before external side effects:

---
model: anthropic:claude-sonnet-4-6
approval: true
channels:
  slack:
    events: [approval]
---

Review pending approvals at http://127.0.0.1:12233/approvals. Approval gates, approval pages, and Slack channels are experimental in this pre-1.0 release.

Works with Anthropic (Claude), OpenAI (GPT), OpenRouter (open source models like GLM and Minimax), and Amazon Bedrock. Switch models with a single line change.

Trigger agents via HTTP webhooks. Integrate with Zapier, Make, GitHub Actions, or any system that can POST. Supports streaming responses for real-time output.

Pause an agent before publishing, sending, deploying, or changing external state. Reviewers can approve, reject, or comment from the web approval dashboard, with optional Slack notifications and threaded review context. Experimental while the API and channel configuration settle.

Schedule agents to run automatically with built-in cron support. Use intervals for sub-daily (5m, 2h) or cron expressions for daily+ (0 9 * * *).

Define agents as .agentuse files with YAML frontmatter and plain English instructions. Version control, code review, and collaborate on agents like any other code.

Connect to any Model Context Protocol server. Access databases, APIs, file systems, and external services through a standardized tool interface.

Compose complex workflows by delegating tasks to specialized child agents. Parent agents can spawn sub-agents with isolated contexts and step limits.

Create reusable agent instructions as SKILL.md files. Reuse your existing Claude Code skills directly - AgentUse reads from the same .claude/skills/ directories. List available skills with agentuse skills.

Full execution history with message logs, tool call traces, token usage, and timing metrics. Debug and audit agent runs with agentuse sessions.

npm install -g agentuse

Set your API key:

agentuse auth login

Full guides and API reference at docs.agentuse.io

• Report bugs
• Share ideas

Apache 2.0