Argentor Documentation
Welcome to the Argentor docs. Argentor is a production-grade autonomous AI agent framework built in Rust — secure by default, composable by design.
v1.3.0 stable — 17 crates · 5,359 tests · 8 LLM backends · AGPL-3.0-only
Why Argentor?
- Security-first: 6-layer guardrails — PII detection, prompt injection, toxicity, shell injection, base64 exfiltration, unicode smuggling — all before any LLM call.
- Minimal overhead: ~2 ms framework overhead vs 11–55 ms for Python alternatives. Measured with paired t-tests at p < 0.0001.
- Cost-optimised: 7.9× fewer tokens on tool-heavy workloads vs LangChain. $491 K/yr savings vs CrewAI at 100 K req/day.
- Composable: 13 focused crates — include only what your agent needs.
- Multi-agent: Orchestrator-Workers pattern, TaskQueue, AgentMonitor, human-in-the-loop.
Quick Navigation
| Goal | Start here |
|---|---|
| First time using Argentor | Getting Started |
| See it running live | Demo |
| Compare to LangChain / CrewAI | Benchmarks |
| Deploy for your organisation | Enterprise |
| Full API surface | docs.rs |
Architecture at a Glance
The workspace contains 13 crates arranged in layers:
argentor-core ← types, errors, event bus argentor-security ← permissions, RBAC, audit, crypto argentor-session ← session management, conversation history argentor-skills ← Skill trait, WASM runtime, marketplace argentor-agent ← AgentRunner, 14 LLM backends, guardrails argentor-channels ← Channel trait (Slack, WebChat) argentor-gateway ← HTTP/WS gateway, REST API, dashboard argentor-builtins ← 50+ built-in skills argentor-memory ← vector store, RAG, embeddings argentor-mcp ← MCP client/server, credential vault argentor-orchestrator ← multi-agent engine, workflows argentor-compliance ← GDPR, ISO 27001, ISO 42001 argentor-cli ← CLI binary + demos
License
All Argentor crates are released under AGPL-3.0-only.