Agent Loop docs

agentloop documentation

agentloop is a synchronous Python agent loop for Anthropic, OpenAI, OpenRouter, Modal/vLLM, and custom clients. It supplies shared conversation types, tool execution, retries, limits, and operational reporting. Your application chooses prompts, tools, approvals, and secrets.

Start here

1. Install agentloop.
2. Run Hello World with the included FakeClient to see the loop work offline, then use the same Agent.run() call with a live provider.
3. Configure a provider with an explicit mapping or Settings object.

Guides

• Agents
• Tools and approval
• Conversations and transcripts
• Providers
• Hosted tools and web search
• Implement a custom provider
• Conversation summaries
• Event traces
• Observers
• OpenTelemetry tracing and metrics
• Installing
• Retries and errors
• Store transcripts

Reference and examples

• API reference
• Example index

The API reference describes the main public interfaces. Names beginning with _ are internal and can change without notice.