Persistent, shared project memory for your AI coding assistants. Open a new AI session and keep going — no re-explaining your architecture, decisions, or progress.
UACE gives Claude Code, Cursor, VS Code Copilot, and any other MCP-capable tool one shared "Project Brain" for each project: architecture and standards, current task and TODOs, recent sessions, git history, and live file activity — all stored locally on your machine and searchable by meaning.
What you get
- Shared memory across tools — a fact saved in one AI tool is available in every other.
- Auto-sync — open a project folder and UACE automatically scans it (languages, frameworks, structure), ingests git history, imports your recent Claude Code sessions, and watches it for changes. No manual setup per project.
- Semantic search — find memories by meaning, not just keywords (local, private, CPU-only embeddings).
- Project Brain sidebar — browse memory, sessions, active files, and commits in the Explorer.
- Works with VS Code Copilot out of the box — the extension registers a local MCP server, so Copilot's agent can read and write your project memory with no config.
- Autonomous mode (just install) — with a one-time consent, UACE auto-registers the
MCP server in your host (VS Code, Cursor, Claude Code, Antigravity) and writes
rules (
AGENTS.md,.cursor/rules, Copilot instructions) + Claude Code hooks so any AI tool recalls context at session start and saves a checkpoint at the end — no manual tool calls. Run “UACE: Remove Autonomy” to cleanly undo every change.
Requirements
- Node.js on your machine (the local memory engine runs on it). UACE auto-detects
Node from your PATH, nvm, fnm, asdf, or system install. If it can't find it, set
uace.nodePathto your Node binary. - VS Code 1.101+ (for the native MCP integration).
On first activation UACE installs its engine locally (one-time, ~1–2 min). After that it starts instantly and works offline.
Use it with Claude Code / Cursor / other tools
Easiest: run “UACE: Set Up Autonomy” — it registers the server for your tools and writes the rules/hooks automatically. Prefer to wire it by hand? Run “UACE: Copy MCP Config for Claude Code / Cursor” for a ready-to-paste snippet that points those tools at the same local memory — so every assistant shares one brain.
Commands
| Command | What it does |
|---|---|
| UACE: Set Up Autonomy | Auto-register the MCP server + write rules/AGENTS.md + Claude hooks |
| UACE: Remove Autonomy | Cleanly remove every UACE-written config entry, rules block, and hook |
| UACE: Sync Current Project Now | Re-scan + import sessions + watch the open folder |
| UACE: Continue Previous Session | Open the full project context packet as markdown |
| UACE: Save Session | Capture a session summary + next steps |
| UACE: Copy MCP Config… | Snippets to connect Claude Code / Cursor |
| UACE: Refresh | Reload the sidebar |
Settings
| Setting | Description |
|---|---|
uace.autoSync |
Auto scan/import/watch the open folder on startup. Default true. |
uace.nodePath |
Override: absolute path to Node (only if auto-detect fails). |
uace.serverPath |
Override: path to a local server build (for development). |
uace.dbPath |
Override: memory database path. Default ~/.uace/memory.db. |
Privacy
Everything is local-first. Your memory lives in a SQLite database at
~/.uace/memory.db. Nothing is sent anywhere; embeddings run on your CPU.
Built on the Model Context Protocol. The engine is
also available standalone on npm as uace-mcp.