Engram

Autonomous agents forget everything between sessions — and worse, they can't tell what they remember isstill true. An agent pulls an old README, a retired price, an API doc you changed months ago, and quotes it with full confidence, because keyword and vector search both rank by resemblance, not truth. Engram makes"is this still true"a first-class, written property: mark a fact superseded or expired and searchwithholds it — and tells the agent what it skipped and why.Per-agent read/write tokens and a git audit trail of who-wrote-what keep it sane when the writers are a fleet, not just you.

Unlike a headless memory store,you can watch it happen.A fast dashboard lets yousearch your brain, see exactly what every agent and teammate changed (with per-filediffs),jump backinto recent notes, and curate it all — while agents read and write the same vault over one MCP endpoint. No database: your.mdfiles are the source of truth, git is the durable store, and an in-memory index powers full-text search + a wikilinkknowledge graph.

Opinionated abouthowit stores memory— git-backed markdown, no database, agents write (not just read), self-hosted.Unopinionated aboutwhatyou keep in it— any markdown vault, any folder structure, any MCP client. Point it at a fresh repo or your existingObsidian vault: no import step, no lock-in.

- Shared memory for a team running multiple agents— one vault, many agents reading and writing concurrently, with per-agent read/write tokens and a git audit trail of who changed what.
- Memory that knows what's still true— retire a price, a term, or a changed API doc and your agents stop quoting it; they're told what they skipped and why. The failure a single agent's memory never fixes.
- Long-term memory for Claude Codeand other coding agents — stop re-explaining your project every session.
- A self-hosted, Obsidian-compatible second brainexposed over MCP — your notes, your server, your git repo.
- Memory you can see, not a black box— a dashboard to search, watch (with diffs), and curate what your agents remember.
- Markdown RAG without the vector database— full-text search + a link graph over human-readable files.

Most agent memory is built to answer"what did I store about this?"Engram is built to answer"what is still true about this?"— a different question, and the one that bites when an agent quotes a price you retired months ago.

The row that matters is the third one. Similarity search cannot tell a contradiction from a duplicate — a retired price and a live one are textually identical, so the retired one oftenoutranksthe live one by being longer and more detailed. That can't be fixed at read time, which is why Engram writes the retirement down when it happens.

This exact pair ships insample-vault/. Runbun dev, searchacme price, and watch the retired note get withheld with a reason.

Categories, not feature-by-feature audits of specific products, and accurate to the best of my knowledge as of July 2026. If something here misrepresents a tool you maintain, open a PR — I'll fix it.

Samebrain_tools. Built fromDockerfile.mcp.

Each mode ships as its own image, and they are not interchangeable:

Deploying the stdio image as a web service is the one mistake worth calling out: it can only 502, because there is nothing listening. It now refuses to start on a PaaS and tells you this instead.

Agents only ever see the active vault — no repo, workspace, or GitHub tools are exposed. Aread-scope token sees only the read tools.brain_captureappears only when the Curator isfull.

Connect an agent (the dashboard →Connectpage shows the exact command + token):

claude mcp add --transport http engram https://<host>/api/mcp \ --header "Authorization: Bearer <token>"

Runs anywhere you can run a Docker container with a persistent volume — Railway, Render, Fly, or your own box.Serverless (Vercel) won't work: Engram holds a volume, a file watcher, and an in-memory index that a serverless function can't keep alive.

- Railway:New Project →Deploy from GitHub repo→ add a Volume at/data.
- Render:one-click via the bundledrender.yaml(Docker + a/datadisk).

Deploying from a prebuilt image instead of the repo? Useghcr.io/rwnalds/engram-app:latest— notengram:latest, which is the stdio server and cannot answer HTTP. Set the healthcheck to/api/health.

How do I give Claude Code long-term memory?Deploy Engram, connect a markdown vault, andclaude mcp addthe endpoint. Thebrain_tools let Claude Code search, read, and write persistent notes across sessions.

How do I stop an agent from quoting outdated facts?Retire the fact and search stops surfacing it. When a value changes, callbrain_supersede(old, new)— one atomic commit marks the old notesuperseded_bythe new one, and it'swithheld from search by default(even if it'slocked). Or setvalid_until: 2026-12-31on a note and it self-expires. Retired matches don't vanish silently:brain_searchreturns them in anexcludedlist with a reason ("expired 2026-06-01"), so the agent can saywhat it ignored and whyinstead of quoting it.

What stops an agent just adding a second, contradicting note?Engram refuses the write. Told "the price is now X", an agent that can't overwrite a note it never read will happilyadd*acme-pricing-2026.mdnext toacme-pricing.md— nothing is corrupted, and you now have two live notes disagreeing about one number. That write is rejected with a pointer tobrain_supersede, which retires the old note and adds the new one in a single commit. Passallow_conflict: truewhen both notes genuinely belong.

How do I know what an agent changed?Every write is committed to git attributed to the token or human behind it, and the dashboard's activity feed shows per-file diffs — a built-in audit trail for autonomous agents.

bun install && bun dev bun test # the ranking/authority suite, incl. the stale-truth fixtures

If Engram is useful to you,starring the repogenuinely helps other people find it.

Next.js 16 (App Router) · React 19 · TypeScript · Tailwind v4 · shadcn/ui · bun · MiniSearch · d3-force · MCP SDK.MIT licensed.

Keywords:MCP server · Model Context Protocol · second brain for AI agents · agent memory · long-term memory for Claude Code · shared memory for AI agents · self-hosted knowledge base · Obsidian-compatible · markdown · knowledge graph · wikilinks · PKM · Zettelkasten · git-backed notes · Hermes agent memory · Cursor memory · RAG without a vector database · chat with your markdown notes · git-backed agent activity feed · audit trail for AI agents · authority-aware search · read-only vs write MCP tokens · agent access control · self-organizing notes · agentic note capture · AI that files your notes · temporal validity · stale memory · agents quoting outdated facts · supersede · note expiry · shared memory for a team of agents · Basic Memory alternative · mem0 alternative.

Creates commit messages from staged files in a local git repository.

Anchor is local repo and org memory for AI coding agents. It indexes GitHub PR history, current code, tests, regressions, architecture, and cross-repo impact locally, then exposes concise cited context through MCP and CLI workflows. Local-first. Read-only GitHub access. No CLI telemetry. No SaaS. No remote LLM calls.

An MCP server that equips your AI coding assistant with live, verified Android knowledge — so it builds from official sources, not from memory.

An MCP server for interacting with the AtomGit API for version control and code hosting.

Design system MCP server — query tokens, components, icons, and WCAG contrast data from Git-backed design systems.

Provides code context from local git repositories.

Local-first documentation for AI agents. Indexes docs from any git repo into SQLite for offline, instant, private access to up-to-date library documentation.

DOS is a small open-source kernel that doesn't believe your AI agents: it verifies what actually shipped from git evidence, arbitrates file collisions between parallel agents, and refuses with structured reasons.

Prevents regression by providing Blast Radius data to AI based on your git history