Memory & Personal Knowledge

Memory is continuity — the thing that makes an actor more than a stateless chat. Hermes keeps MEMORY.md and USER.md, indexes sessions with SQLite + FTS5 for search, and distinguishes what's a durable fact from what's just operational state. The hard part is signal: remembering enough to be useful without becoming noisy.

The goal is an agent that remembers my VDP, my atlases, my stack, and my preferences without drowning in trivia. And when several actors need a shared model of "who I am", Honcho gives each profile its own AI peer over a common user workspace — the answer to "isolated actors that still need one model of me".

Planned notes

• MEMORY.md: the agent's working memory file
• USER.md: the durable model of me
• Session search over SQLite + FTS5
• Factual memory vs operational memory
• Consolidation: turning sessions into durable notes
• Pruning: what to forget, and when
• What to keep vs what's just noise
• Honcho: per-profile AI peer over a shared workspace
• The multi-agent shared-memory model
• Keeping memory in sync with the atlases and VDP

Core sources

• Hermes — Feature: Memory — https://hermes-agent.nousresearch.com/docs/user-guide/features/memory
• Hermes — Feature: Memory Providers — https://hermes-agent.nousresearch.com/docs/user-guide/features/memory-providers
• Hermes — Session Storage (FTS5) — https://hermes-agent.nousresearch.com/docs/developer-guide/session-storage
• Honcho — shared multi-agent memory layer — TODO confirm canonical URL (see SOURCES.md).

Connects to: Hermes Architecture · Identity & Context · Profiles, Actors & Delegation · Skills & Procedural Memory