Persistent AI memory
Add a memory layer so AI agents keep their context after the session ends.
SYNOPSIS
memory [--remember] [--recall] [--reflect] [--forget]DESCRIPTION
An AI agent forgets its decisions and dead ends when the session ends. Keep memory on a server outside the agent and it carries over across sessions and tools.
Based on AnchorMind, which we build and run ourselves, we set up and connect the memory layer to fit your environment.
SCOPE
- Memory model
- Facts, decisions, errors, preferences, procedures, relations, and episodes are stored separately so they can be recalled precisely.
- Search
- Keyword, semantic, and hybrid layers, with a warning on stale memories.
- Access control
- Per-agent tokens and workspace isolation keep memories from mixing.
- Integration
- Connect to agents such as Claude Code and Codex over MCP, with session start and end hooks.
DELIVERABLES
- Memory server setup
- Agent integration code and hooks
- Store and recall rules document
- Operations guide
FIT
Good fit
- You use several AI tools and keep re-explaining the same context
- You want agents to consult a person’s or team’s decision history
Not a fit
- You only need one-off questions and answers with no history