CrossAgentMemory 🧠

Open-source cross-agent memory layer for AI coding agents.

Your AI agent should remember what you built yesterday, why you rejected that approach last week, and that you prefer async/await over callbacks.

CrossAgentMemory makes that happen.

The Problem

Every time you start a new session with Claude Code, Codex, or Cursor, your agent remembers nothing. You re-explain your codebase. You re-teach your preferences. You burn tokens and patience.

CrossAgentMemory fixes session amnesia.

Features

Install

pip install crossagentmemory
# or
pipx install crossagentmemory
# or
uv tool install crossagentmemory

Quick Start

1. Initialize memory for your project

cd my-project
crossagentmemory init

2. Capture memories as you work

crossagentmemory capture "Chose PostgreSQL over MongoDB for ACID compliance" --category decision --confidence 0.95

crossagentmemory capture "Always use async/await, never callbacks" --category preference

crossagentmemory capture "Auth bug: JWT refresh tokens not rotating" --category error

3. Find related memories (semantic search)

crossagentmemory related "authentication flow"

Output:

┌────┬────────────┬──────────┬────────────────────────────────────────┐
│ ID │ Similarity │ Category │ Content                                │
├────┼────────────┼──────────┼────────────────────────────────────────┤
│ 4  │ 0.58       │ decision │ Chose FastAPI over Django for async    │
└────┴────────────┴──────────┴────────────────────────────────────────┘

4. Summarize your project

crossagentmemory summarize

5. Sync to CLAUDE.md

crossagentmemory sync

This generates CLAUDE.md in your project root. Claude Code reads it automatically on startup.

6. Install git hooks (auto-sync on commit)

crossagentmemory hook install

Now CLAUDE.md updates automatically every time you commit.

Full CLI Reference

# Core memory operations
crossagentmemory init                          # Initialize memory for project
crossagentmemory capture "content"             # Store a memory
crossagentmemory capture "..." --auto-tag      # Auto-generate tags with LLM
crossagentmemory recall                        # List recent memories
crossagentmemory search "keyword"              # Keyword search
crossagentmemory related "query"               # Semantic search
crossagentmemory summarize                     # Auto-summarize project
crossagentmemory summarize --llm               # LLM-powered rich summary
crossagentmemory summarize --session ID        # Summarize one session
crossagentmemory digest                        # Weekly digest
crossagentmemory digest --llm                  # LLM-powered weekly digest
crossagentmemory check-conflicts               # Detect contradictory memories
crossagentmemory reinforce <id>                # Boost memory confidence
crossagentmemory decay                         # Apply confidence decay
crossagentmemory decay --dry-run               # Preview decay

# Auto-capture
crossagentmemory capture-auto                  # Auto-capture from git + shell + Claude
crossagentmemory capture-auto --dry-run        # Preview what would be captured
crossagentmemory capture-auto --sources git    # Only capture from git log

# Agent integration
crossagentmemory load                          # Generate context brief
crossagentmemory sync                          # Sync to CLAUDE.md
crossagentmemory export                        # Export to markdown

# Team sync
crossagentmemory team export                   # Export memories to .crossagentmemory/
crossagentmemory team import                   # Import team-shared memories
crossagentmemory team status                   # Show team sync status

# Shell integration
crossagentmemory shell show                    # Show shell integration script

# Background daemon
crossagentmemory daemon start                  # Start silent auto-capture
crossagentmemory daemon status                 # Check daemon status

# Import & migration
crossagentmemory import <path> --format mem0   # Import from Mem0
crossagentmemory import <path> --format markdown
crossagentmemory import <path> --format json

# Social sharing
crossagentmemory post "Milestone!"             # Post to Twitter/LinkedIn

# Memory graph
crossagentmemory graph                         # Build relationship graph

# MCP server & dashboard & API
crossagentmemory mcp                           # Start MCP server (stdio)
crossagentmemory dashboard                     # Start web dashboard on :8745
crossagentmemory server                        # Start REST API on :8746

# Backup & restore
crossagentmemory backup                        # Create dated .zip backup
crossagentmemory backup -p my-project          # Backup single project
crossagentmemory restore backup.zip            # Restore from backup
crossagentmemory restore backup.zip --dry-run  # Preview restore

# Management
crossagentmemory stats                         # Show statistics
crossagentmemory hook install                  # Install git hooks
crossagentmemory hook uninstall                # Remove git hooks
crossagentmemory delete <project>              # Wipe project memory

How It Works

Your Terminal Agent
       │
       ├──► crossagentmemory capture "..."   ──► SQLite (~/.crossagentmemory/memory.db)
       │
       ├──► crossagentmemory capture-auto    ──► Auto-import from git / shell / Claude
       │
       ├──► crossagentmemory related "..."   ──► TF-IDF + Cosine Similarity (numpy)
       │
       ├──► crossagentmemory load            ──► Markdown brief for agent context
       │
       ├──► crossagentmemory sync            ──► CLAUDE.md (auto-read by Claude Code)
       │
       ├──► crossagentmemory team export     ──► .crossagentmemory/ (git-shared)
       │
       ├──► crossagentmemory mcp             ──► MCP server for Cursor/Copilot/Claude
       │
       └──► crossagentmemory dashboard       ──► Web UI on http://localhost:8745

Storage: Plain SQLite. Query it with any tool. Back it up. Version it.

Semantic Search: Built with TF-IDF + cosine similarity in pure numpy. No scikit-learn, no sentence-transformers, no API calls.

Privacy: Nothing leaves your machine.

Memory Categories

Why Not Just Use CLAUDE.md?

CLAUDE.md is static documentation. CrossAgentMemory is learned memory:

Use both. CLAUDE.md for stable conventions. CrossAgentMemory for dynamic context.

Storage Backends

CrossAgentMemory supports multiple storage backends. SQLite is the default — zero config, works offline, perfect for individuals.

PostgreSQL is available for teams, concurrent access, and larger deployments.

SQLite (default)

# No configuration needed — works out of the box
crossagentmemory init

Database file: ~/.crossagentmemory/memory.db

PostgreSQL

# 1. Install with PostgreSQL support
pip install crossagentmemory[postgres]

# 2. Start PostgreSQL (Docker Compose included)
docker compose up -d

# 3. Set the connection URL
export DATABASE_URL=postgresql://crossagentmemory:crossagentmemory@localhost:5432/crossagentmemory

# 4. Initialize
crossagentmemory init

Switching Backends

from crossagentmemory import MemoryEngine

# Auto-detect: uses Postgres if DATABASE_URL is set, otherwise SQLite
engine = MemoryEngine()

# Explicit SQLite
engine = MemoryEngine(backend="sqlite")

# Explicit PostgreSQL
engine = MemoryEngine(backend="postgres")

Migrating Data

# Migrate all memories from SQLite to PostgreSQL
crossagentmemory migrate --from-backend sqlite --to-backend postgres

# Migrate a single project
crossagentmemory migrate -p my-project --from-backend sqlite --to-backend postgres

# Specify custom source DB or target DSN
crossagentmemory migrate --from-db-path ./old.db --to-dsn postgresql://user:pass@host/db

Roadmap

• Core SQLite engine
• CLI (init, capture, recall, search, load, export)
• Markdown context brief generation
• CLAUDE.md sync
• Git hooks
• Semantic search (TF-IDF + cosine similarity)
• Auto-summarization
• Confidence decay & reinforcement
• Auto-capture from agent sessions
• Team sync (shared memory via git)
• MCP server integration
• Web dashboard
• Shell integration
• Background daemon
• Memory graph visualization
• Mem0 importer
• Social sharing
• VS Code extension
• Pluggable storage backends (SQLite + PostgreSQL)
• Schema versioning
• Backup & restore
• Config file support
• FTS5 full-text search

Contributing

PRs welcome! This is a community project.

git clone https://github.com/Vish150988/crossagentmemory.git
cd crossagentmemory
pip install -e ".[dev]"
pytest tests/ -v

License

MIT © Open Source Community