Quick Start

IntroductionCopy Anchor Link

Tiramisu is an open source git-native memory system for AI coding agents. It stores curated memories as Markdown files inside your repositories, versioned alongside your code and reviewable in pull requests.

FeaturesCopy Anchor Link

• Hierarchical memory scoping
• Separate project, personal, team, and organization memories
• Easy pruning
• Custom memory schemas and writing rules
• Supports many languages and agent harnesses

InstallationCopy Anchor Link

Step 1: Configure pruning (Optional)Copy Anchor Link

Upvotes and pruning reduce stale memories.

When a memory helps solve a task, it gets upvoted, which extends its lifespan. These events are stored in the database. On request, the agent can prune memories, meaning it reviews expired memories and suggests updates or deletions.

If you skip this step, upvotes are disabled, but the agent can still prune manually by searching your codebase to verify relevance.

Option A: Self-HostedCopy Anchor Link

Spin up a PostgreSQL database to store upvotes.

Option B: Tiramisu app (Coming Soon)Copy Anchor Link

• ⚡ Cloud Database
• 📊 Dashboard
• 🤖 Automated Pruning PRs

Step 2: Separate project memories from personal, team, and/or organization memories (Optional)Copy Anchor Link

This step is useful for teams. Solo devs can skip to Step 3.

1. Create one repo per set of memories you want to keep separate (e.g. per team, one company-wide, one personal).
2. Add each repo to your IDE and agent harness workspaces to "import" the memories.
3. Install Tiramisu in each repo:

npx tiramisu@latest init --availableToWorkspace

Step 3: Install to projectCopy Anchor Link

Inside your project(s), run:

npx tiramisu@latest init

Manual InstallationCopy Anchor Link

1. Install the tiramisu CLI:

npm i -g tiramisu

2. Create tiramisu.json at the root of your repository with the options you want:

{
  "version": 1,
  "availableToWorkspace": false,
  "prune": {
    "unvotedTtl": "90d",
    "humanUpvoteTtl": "180d",
    "agentUpvoteTtl": "90d",
    "databaseUrlCommand": "doppler secrets get TIRAMISU_DATABASE_URL --plain"
  }
}

3. Add the MCP tools:

Option 1: Using add-mcp

npx add-mcp "tiramisu mcp" -g --auto-approve

Option 2: Manual installation

Refer to your agent's documentation on how to add a global MCP server.

Add:

tiramisu mcp

4. Add the memory writing skill:

npx skills add buildsip/tiramisu --global

5. VS Code / Cursor memory tab labels

{
  "workbench.editor.customLabels.patterns": {
    "**/.memories/**/memory.md": "${dirname}/memory.md"
  }
}

6. If you've enabled pruning, migrate your database using the drizzle schema from tiramisu/packages/cli/migrations.