LegionForge Framework¶
The hardened, self-hosted alternative to cloud agent platforms.
LegionForge is an open-source framework for running AI agent systems on your own hardware. It runs local LLMs via Ollama or cloud APIs (OpenAI, Anthropic), with a full security stack baked into every layer of the execution pipeline — not bolted on afterward.
github.com/LegionForge/LegionForge
Why it exists¶
In January 2026, OpenClaw hit 60,000 GitHub stars in 72 hours and 300,000+ users in weeks. Kaspersky found 512 vulnerabilities (8 critical). Cisco found active data exfiltration in third-party skills. LegionForge is built in the opposite order: security first, product on top.
What's in the box¶
- Local-first LangGraph runtime — agents execute on your hardware; cloud APIs are optional
- Deterministic security pipeline — prompt-injection detection, tool-revocation, capability-boundary checks, Ed25519 tool signing, adaptive threat rules
- Three independent loop-protection layers — step counter, action-history loop detection, token-budget guard
- Multi-provider LLM factory — Ollama, OpenAI, Anthropic, InceptionLabs (mercury-2) behind one interface
- PostgreSQL state layer — async pool, LangGraph checkpoint resumption, pgvector RAG, threat-event audit chain
- Gateway API — FastAPI with task queue, SSE streaming, A2A + MCP endpoints, Bearer auth
- Connectors — Discord, Telegram, Slack, WhatsApp, webhooks
- Web UI — task submission, session continuity, multi-modal image input, HITL approval gate
- Guardian sidecar — 7-check deterministic pipeline running as a separate process
Pages in this section¶
-
Install, set up the venv, configure the hardware profile, run a smoke test, send your first task through the gateway.
-
Core design principles. Module map. How a task flows from gateway → orchestrator → agents → tools → response.
-
Where validation happens, what threats it catches, how Guardian fits in.
-
The hardware profile system. Pydantic settings singleton. Switching between profiles.
-
PostgreSQL schema, role separation, LangGraph checkpoints, pgvector RAG, audit chain.
-
Task submission, SSE streaming, A2A + MCP endpoints, auth model.
-
Discord, Telegram, Slack, WhatsApp, webhooks. How a bot bridges to the gateway.
-
Structured event types logged to
threat_events. What each one means and what fires it. -
Auto-generated reference from the framework's Python source. Coming soon.