Documentation
Build, configure, and orchestrate secure agents.
Everything you need to install RavenClaws, point it at a provider, wire up tools, and scale from a single agent to an autonomous swarm.
Getting started
Install via Cargo, Docker, or source. Configure a provider and run your first one-shot task, REPL session, and a multi-step task with tools.
Configuration reference
The full layered config system: file → env → flags. Every section, key, default, and CLI flag — with example configs for each mode.
Swarm mode
Multi-agent orchestration: flat swarms, hierarchical supervisors, self-provisioning workers, the inter-agent message bus, and health monitoring.
MCP integration
Use RavenClaws as an MCP client to consume external tools, or as an MCP server to expose its own — all policy-checked and audited.
Heartbeat mode
Autonomous, long-running agents on an assess → plan → act → persist → sleep cycle. State survives restarts, so agents resume from the last checkpoint and run unattended for days.
Server mode
Run RavenClaws as an HTTP server with a REST API — chat, execute tasks, list and invoke tools, health checks, Prometheus metrics, and SIGHUP config reload.
vLLM integration
High-throughput, GPU-accelerated local inference via the OpenAI-compatible provider. Quick start, configuration, and tool-calling support.
llama.cpp integration
Lightweight, CPU-first local inference with GGUF models. Quick start, configuration, performance tips, and text-based tool-calling fallback.
Migration guide
Breaking changes and migration paths between RavenClaws versions — from v0.3 through v1.0. Covers API changes, removed types, and new config sections.
These guides mirror the docs/guides/ directory in the repository. For the complete generated API, see the rustdoc on docs.rs, and the runnable examples on GitHub.