Power your agent
คำสั่งของผู้เผยแพร่เอง ไม่ใช่เอนด์พอยต์ของ PlugMCP
Host a copy with our keys and logs.
npx -y @aictx/[email protected]Claude Code ไม่มีลิงก์ติดตั้งสำหรับคำสั่งแบบโลคัล ให้คัดลอกคำสั่งด้านล่างแล้วรันในเทอร์มินัล
claude mcp add aictx-memory -- npx -y @aictx/memory@0.2.1รันในเทอร์มินัล
{ "mcpServers": { "aictx-memory": { "command": "npx", "args": [ "-y", "@aictx/[email protected]" ] } }}claude_desktop_config.json
URL แบบรีโมตหรือคำสั่งบนเครื่องของคุณ? การติดตั้งสองแบบต่างกันอย่างไร
รันบน PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
เปิดจากผู้เผยแพร่
Your agent knows your code. It doesn't know your product.
Memory keeps the product graph of a codebase: features with their lifecycle stage, decisions with their reasons, gotchas, and open questions — anchored to the code paths they describe. Agents query it on demand and keep it current as the product changes.
Local-first product graph for AI coding agents.
If you run several projects with AI coding agents, you know the failure mode. You come back to a repo after three weeks. Neither you nor the agent remembers what is shipped, what is half-done, why the queue lives in the worker, or what was decided about storage. The agent can re-derive code structure cheaply; it cannot re-derive intent, stage, or rationale, because that knowledge is not in the repo. So you re-explain, or the agent guesses.
Memory stores that product layer as plain local files in .memory/: JSON
sidecars with Markdown bodies, indexed with SQLite full-text search. No
embeddings, no cloud account, no model API. Saved memory is reviewable in Git
like any other change. Code-graph tools map what is in the repo; Memory
captures what isn't.
This repository publishes the npm package @aictx/memory and the Homebrew
formula aictx/tap/memory. It ships the memory CLI and the optional
memory-mcp server.
init once -> query on demand -> save product changes -> sync at session end
memory init — once per repo. Creates .memory/ storage, installs a
short guidance block plus an auto-generated product map into AGENTS.md
and CLAUDE.md (marker sections), starts the local viewer, and prints an
indexing brief. Your coding agent follows the brief to build the initial
graph: explore the repo, interview you for intent, stage, and decisions,
then save everything in one call.memory query "<question>" — on demand, mid-task. Returns a
token-budgeted Markdown subgraph: full-text matches, their one-hop
relations, and connected open questions. There is no per-task context
loading; the only always-on context is the roughly one-screen product map.memory save --stdin — after product-meaningful changes. The input is
intent JSON: {task, nodes, stale, supersede, delete}. The product map in
AGENTS.md/CLAUDE.md refreshes automatically on every save.memory sync — at session end or after merging. Diff-driven: verifies
anchors against the actual tree, reports nodes whose anchored code changed,
orphaned anchors, and coverage gaps, then prints an agent prompt with a
pre-filled save skeleton. Sync is mechanical — it never writes graph nodes
itself.memory status — features by stage, open questions, stale anchors,
last activity and sync. memory status --all is the cross-project
dashboard: one row per registered project, for people running many repos
with agents.The map is the only thing agents carry into every session. It is generated,
capped at roughly one screen, and refreshed by save and sync:
## Product map (generated — do not edit; refresh with memory save or memory sync)
Acme Render — Self-hosted PDF render service for invoice pipelines.
**Building:** webhook-retry-queue — Failed Stripe webhooks re-enter a worker-owned retry queue. — services/billing/src/webhooks/
**Shipped:** pdf-render-api — Renders invoice templates to PDF over HTTP. — services/render/ · template-editor — In-browser template editing with live preview. — apps/editor/
**Paused:** batch-exports — Nightly bulk export of rendered documents. — services/exports/
**Recent decisions:** retries-run-in-worker — Webhook retries run in the queue worker · sqlite-over-postgres — SQLite for single-node deployments
**Open questions:** pdf-storage-location — Where do rendered PDFs live long-term?
Everything deeper is pull, not push: memory query when the agent needs it.
Memory requires Node.js >=22. The Homebrew formula installs Node through
Homebrew; npm installs require a compatible Node already on PATH. Core
commands run locally; no cloud account, model API, embeddings, or hosted sync
are required.
# macOS/Linux with Homebrew
brew install aictx/tap/memory
# or npm
npm install -g @aictx/memory
cd path/to/your/repo
memory init
memory init creates local storage, installs the guidance block and product
map sections, starts the viewer, and prints the indexing brief for your agent.
Use memory init --no-view to skip the viewer, memory init --dry-run to
preview, or memory init --brief to reprint the brief later.
Memory writes local files and never commits automatically.
Paste this into Claude Code, Codex, OpenCode, Cursor, Cline, or another CLI-capable coding agent from the project root:
Set up Memory for this repository.
Install Memory with one of:
brew install aictx/tap/memory
npm install -g @aictx/memory
Then run from the project root:
memory init
Follow the printed indexing brief: explore the repo, draft the product graph
(features with stage and anchors, decisions with their reasons, gotchas, open
questions), interview me for what the repo cannot tell you, and save it all in
one `memory save --stdin` call.
When this is done, report:
- what the product map in AGENTS.md/CLAUDE.md now shows
- what `memory status` reports
- how I can inspect the graph with `memory view`
After the initial graph exists, the ongoing loop is small:
memory query "why do webhook retries run in the worker?"
# do the work
memory save --stdin
memory sync # at session end
Five kinds of node, four relation predicates.
Nodes connect through affects, depends_on, supersedes, and related_to
relations. Anchors are what make staleness mechanical: when anchored files
change or disappear, sync and status report exactly which nodes need
re-verification.
What this buys you, concretely:
memory sync is how the graph stays honest without re-reading the repo:
memory sync
It diffs the tree since the last sync marker (.memory/sync-state.json),
checks every anchor against the current files, and reports four things: nodes
whose anchored code changed, nodes whose anchors match nothing anymore, nodes
with no anchors, and directories with real code but no feature coverage. When
something needs attention it prints an agent prompt with a pre-filled
memory save --stdin skeleton, so reconciliation is one save call. Run it at
session end or after merging others' work. --dry-run reports without
advancing the marker.
memory view starts a local browser viewer bound to 127.0.0.1: a projects
dashboard, the memory list, node detail, and an interactive relation graph.
memory inspect <id> does the same for one node in the terminal, and
memory diff shows tracked and untracked .memory/ changes.
Codex users can add this repo's plugin marketplace with one command:
codex plugin marketplace add aictx/memory
Claude Code users can add the marketplace and install the plugin from inside Claude Code:
/plugin marketplace add aictx/memory
/plugin install memory@aictx
For official listing paths and release prep, see Publishing agent plugins.
Memory is pre-1.0. Breaking changes should be expected across package versions
while the schema and local storage format are still evolving. There is no
storage migration: when a new version rejects existing .memory/ storage,
reset and re-index from the project root:
memory reset
memory init
memory reset creates a backup archive under .memory/.backup/ before
clearing local storage. To rebuild the graph afterwards, ask your coding agent:
Memory was reset after a package upgrade. Run `memory init` and follow the
indexing brief to rebuild the product graph. Inspect the newest archive in
.memory/.backup/ for prior features, decisions, gotchas, and open questions,
and re-save only what is still true in one `memory save --stdin` call. Do not
copy old storage files back; validate the result with `memory check`.
Memory is MIT-licensed and built in the open. Issues, docs fixes, examples, and pull requests are welcome.
Memory by Aictx keeps the product graph of a codebase for AI coding agents:
features with stage and code anchors, decisions with their reasons, gotchas,
and open questions, stored as reviewable local files. It is distributed
through the open source npm package @aictx/memory and the Homebrew formula
aictx/tap/memory, then runs through the memory CLI and optional
memory-mcp server.
{ "mcpServers": { "aictx-memory": { "command": "npx", "args": [ "-y", "@aictx/[email protected]" ] } }}เครื่องมือ
ยังไม่มีรายการเครื่องมือ เซิร์ฟเวอร์แบบรีโมตจะได้รายการหลังการตรวจสอบสถานะครั้งแรก
FAQ
คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
@aictx/memory ทำงานแบบโลคัลผ่าน stdio
ตรวจสอบล่าสุดเมื่อ 2026-10-28 Local-first product graph for AI coding agents.