Concepts

A soft intro to the ideas in LegionForge before you dive into Architecture, Guardian, or Security. If you've used another agent framework (LangChain, AutoGen, CrewAI, OpenClaw, Hermes, row-bot), most of this will feel familiar. The places it doesn't will be the places LegionForge differs intentionally.

Pages in this section

• Agents and tasks

  What an agent is in LegionForge. What a task is. The lifecycle from submission to result.

• Tools and capabilities

  What a tool is. What capability scope means. Why tools are signed, registered, and gated.

• Security fundamentals

  Trust boundaries. Sanitization. Deterministic checks. The thesis that the LLM is not trustworthy.

• Glossary

  One-line definitions for every LegionForge-specific term — Guardian, HITL, capability scope, threat event, trust boundary, sanitization, A2A, MCP, and the rest.

Where this fits

Concepts is the mental model layer. The order most people benefit from reading:

1. Concepts (you are here) — get the vocabulary
2. Quickstart — run something
3. Framework → Architecture — understand how the pieces compose
4. Security Model or Guardian — go deep on what's different

Architecture is dense. If you read it first without the vocabulary, you'll bounce. Concepts gives you the words to make Architecture land.