What is OpenLegion?
OpenLegion is a container-isolated multi-agent runtime. Every agent runs in its own Docker container with private memory, tools, schedule, and budget — coordinated through a SQLite-backed blackboard, pub/sub events, and a structured handoff protocol. Fleet model — no CEO agent. API keys never leave the mesh — agents proxy LLM calls through a credential vault and never see secrets. Chat with your fleet via Telegram, Discord, Slack, WhatsApp, Webhook, or the CLI REPL. Monitor everything from the engine dashboard (Chat / Work / Team / Settings).
No LangChain. No Redis. No Kubernetes. No CEO agent. 100+ LLM providers via LiteLLM.
Get started
Managed Hosting
Sign up at app.openlegion.ai and get a dedicated agent fleet running in ~5-12 minutes. No infrastructure to manage.
Self-Hosted
Clone the repo and run on your own machine. Full control, no account needed.
Why OpenLegion?
Security by Architecture
Agents run as UID 1000 in cap-dropped, read-only Docker containers with 384 MB RAM / 0.15 CPU defaults. API keys live in a mesh-tier credential vault — agents never see them. Defense-in-depth across container isolation, vault proxy, SSRF guard, AST-validated skills, per-agent budgets, and rate-limit categories.
Fleet Model — No CEO Agent
Agents coordinate through a SQLite-backed blackboard (with atomic CAS), pub/sub events, per-agent FIFO lanes, and a structured handoff protocol. No LLM “router” decides what runs next.
Cost Controls
Per-agent and per-project LLM cost ledger enforces daily (200 default) USD caps before each request. CAPTCHA spend ledger tracked separately in millicents.
Self-Improving Agents
Agents learn from tool failures and user corrections —
errors.md and corrections.md auto-inject each session. Agents can write their own Python skills at runtime (AST-validated against 23 forbidden imports / 16 forbidden calls / 11 forbidden attrs, max 10K chars) and hot-reload them.Multi-Channel
Connect agents to Telegram, Discord, Slack, WhatsApp, Webhooks, and the CLI REPL. Engine dashboard provides chat, kanban, agent grid, and settings surfaces.
100+ LLM Providers
Routed via LiteLLM (100+ providers). 15 system-tier providers are natively supported with mesh-managed keys; failover chains with per-model health tracking and exponential cooldown.
Explore the docs
Managed Hosting
Dedicated VPS, custom subdomain, automatic provisioning, and subscription management.
Architecture
Trust zones, mesh host, agent containers, and the browser service.
Coordination
Blackboard, pub/sub, lanes, handoffs, and fleet templates.
Configuration
Agents, permissions, mesh settings, and environment variables.