Verbative Memory
On-device memory system for AI coding agents — a dual-layer engine (distilled facts + a
lossless event ledger), automatically captured and injected, exposed over MCP. Shared
across agents, persistent across sessions; stored as plain, human-readable files in your repo —
no memory cloud, no vendor database.
This repository is the public home for docs, install help, and issues. The engine is
closed-source and ships as a sealed, compiled package — there is no source code here.
How it works — capture everything, recall selectively
- Complete capture, in two layers. Underneath sits a lossless ledger: a mechanical,
no-model record of every command, edit, and test result, so "what did we run" and "when did
this test start failing" are exact, replayable facts. On top, an on-device model distills that
stream into atomic, typed memories — decisions, conventions, and dead ends.
- Selective recall — the two layers fused. A single recall surfaces only the handful of
distilled facts that matter (ranked by semantic search and a reranker) and interleaves them,
by time, with the exact ledger record around them: what you decided next to what you
actually did. One answer, not a dump of the whole history.
- Self-organizing. New facts reconcile against old ones, importance decays on per-type
forgetting curves, frequently-recalled facts strengthen, and recurring specifics distill into
higher-level insights.
- Yours, on-device. Memories live in human-readable files in your project. Recall makes
zero cloud calls.
Requirements
- macOS (Apple silicon) or Windows 10/11 (x64). (Linux is on the roadmap.)
- A Verbative account — Memory is part of the Code Advanced plan and starts with a
14-day free trial. See the plans and the
Terms.
Install
Claude Code CLI — automatic capture and recall:
claude mcp add verbative-memory -- npx -y verbative-memory
pip:
pip install verbative-memory
verbative-memory login
Any MCP client (mcp.json / stdio):
{ "mcpServers": { "verbative-memory": { "command": "npx", "args": ["-y", "verbative-memory"] } } }
Automatic capture and injection are built for the Claude Code CLI and Codex; any other
Model Context Protocol client can call the tools manually. The first run downloads the on-device
engine and models once; after that, recall runs fully offline.
More detail in docs/INSTALL.md and at
verbative.de/docs/memory.
remember · recall · search · ask · list · get · pin · forget · revise ·
edit_core · feedback · related · history · timeline · events — and more. See the
docs.
Privacy
Capture, embeddings, reranking, and the knowledge graph all run locally. Your code and memories
are not uploaded; the memory files stay in your project and leave your machine only if you
choose to share them.
Links
Issues
Bug reports and feature requests are welcome in the issue tracker — the place
for anything about the docs or the tool. For your account, licensing, or a trial, email
info@verbative.de.
Verbative is a product of Cramer Digital. Not affiliated with, or endorsed by, Anthropic.
Verbative works alongside Anthropic's Claude and the Claude Code CLI; those names belong to
Anthropic. See LICENSE.