Agentmem
Persist and govern long-lived facts for coding agents per project with trust scoring instead of stuffing everything into the chat window.
Overview
agentmem is a MCP server for the Build phase that stores governed, project-namespaced agent memory in SQLite with trust and conflict controls.
What is this MCP server?
- Governed memory store with trust lifecycle, conflict detection, and health scoring
- SQLite database via required --db path and --project namespace arguments
- PyPI package quilmem served with positional serve subcommand (v0.2.5)
- stdio MCP transport for local agent hosts
- Designed for coding agents that need durable, scoped recall across sessions
- Registry version 0.2.5
- Required CLI args: --db, --project
- Transport: stdio; registry identifier: quilmem (PyPI)
Community signal: 2 GitHub stars.
What problem does it solve?
Coding agents forget decisions between sessions and silently contradict earlier facts unless you re-paste context every time.
Who is it for?
Solo builders running long-horizon agent sessions on one or few codebases who want local, namespaced memory with governance hooks.
Skip if: Teams that need a hosted multi-user knowledge base, or builders who only need ephemeral chat with no persistence.
What do I get? / Deliverables
Agents read and write durable, scored memories per project so recall stays consistent and conflicts surface instead of hiding in chat.
- Running stdio MCP memory service backed by a local SQLite file
- Project-scoped memory entries agents can query and update with governance metadata
Recommended MCP Servers
Journey fit
Memory infrastructure is shelved under Build because you wire it while setting up agent tooling, even though recalled memories help across later iterate and review work. agent-tooling is the canonical shelf for SQLite-backed MCP memory with project namespaces and lifecycle rules.
How it compares
Governed agent-memory MCP, not a vector RAG pipeline skill or a plain notes app integration.
Common Questions / FAQ
Who is Agentmem for?
Indie developers and solo builders using coding agents who want structured, project-scoped memory with trust and conflict handling.
When should I use Agentmem?
When you start a serious agent workflow on a repo and need memories to survive sessions across Build, review, and iterate without manual re-briefing.
How do I add Agentmem to my agent?
Install quilmem from PyPI, run serve with --db (SQLite path) and --project (namespace), and add that stdio command to your MCP host configuration.