Concepts

The core ideas behind Agent Smith.

• Pipeline system: commands, handlers, presets, and how they chain together
• Specs & runs: how Agent Smith tracks its own work, from planning to execution and results
• Self-documentation: specs, runs and decisions as an audit trail
• Decision logging: why the agent made each choice, not just what it did
• Cost tracking: token usage, pricing, and result.md frontmatter
• Interactive dialogue: structured Q&A between agent and human across all channels
• Knowledge base: an LLM-compiled project wiki from accumulated run records
• Memory: facts the agent keeps across runs