Collective Agent Memory

█▀▀ ▄▀█ █▀▄▀█            🇨​🇴​🇱​🇱​🇪​🇨​🇹​🇮​🇻​🇪​
█▄▄ █▀█ █░▀░█            🇦​🇬​🇪​🇳​🇹​ 🇲​🇪​🇲​🇴​🇷​🇾​

---

Collective Agent Memory

Shared, searchable agent memory across machines

CAM indexes sessions from Claude Code, Cursor, OpenClaw, and Codex CLI, segments them into searchable topics, and optionally syncs them across machines via GitHub. Your agents can search previous work, decisions, and code patterns from any session on any machine.

What It Does

• Context recovery - Continue previous work with one command: cam context "topic"
• Search past sessions - Find previous work with fast weighted keyword search
• Entity search - Find sessions mentioning specific tools, files, concepts, or people
• Filter by time - Get only recent results: last 2 hours, last day, last week
• Filter by agent - Search only Claude, OpenClaw, or other specific agents
• Shared memory - Sync sessions across machines so all agents can search everything
• Topic segmentation - Sessions are automatically split into coherent topics with keywords
• Entity extraction - Typed entities (tools, files, concepts, etc.) are extracted and indexed
• Fast local search - SQLite FTS5, no embeddings needed, sub-second results

Install

curl -fsSL https://raw.githubusercontent.com/julianfleck/collective-agent-memory/main/install.sh | bash

The installer will:

1. Install CAM and its dependencies
2. Detect your host RAM and prompt for local or headed mode (see below)
3. Download ML models (~1.5GB) for embeddings and entity extraction (local mode only)
4. Detect your coding agents (Claude Code, Cursor, OpenClaw, Codex)
5. Index existing sessions into searchable session segments
6. Optionally set up GitHub sync for cross-machine access
7. Install the /cam skill to your agents

**Tip:** Set `HF_TOKEN` before installing for faster model downloads (see [Configuration](#configuration)). The installer + `cam init` writes t