- × Re-explain project context every session
- × Agents duplicate or overwrite each other’s work
- × Decisions disappear into chat history
- × “Done” is hard to verify later
Your coding agents forget.
Your project doesn’t have to.
Attacca OS gives humans and AI coding tools one durable workspace for project memory, tasks, decisions and messages — so context survives the session that created it.
Context shouldn’t reset when the chat does.
AI coding tools are fast inside a session. The hard part is keeping decisions, ownership and project state intact across new sessions, different tools and multiple agents.
- ✓ Durable project memory across sessions
- ✓ Explicit tasks, claims and file scope
- ✓ Decisions and evidence stay attached to work
- ✓ Humans and agents share one workspace
Project memory that behaves like infrastructure.
Not another prompt file. Attacca keeps structured, auditable state that coding tools can read and update as the project moves.
One shared memory
A hash-chained append-only event ledger plus a readable activity log preserves what changed, why it changed and who did it.
Cold resume
Return later and restore the project handoff, shared context, role rules and exact agent identity instead of rebuilding context from scratch.
Coordinated work
Task boards, expiring claims, declared file scope, immutable plan revisions and evidence-based completion keep parallel workers aligned.
Identity stays distinct
AI actor, human operator, runtime, role and Git checkout remain separate audit fields instead of being blurred into one session identity.
Built for the tools you use
Native plugins for Claude Code, Codex CLI and Kimi Code. MCP for Cline, Cursor, Windsurf and other clients. REST and CLI for everything else.
Self-hosted by default
One Python 3.8+ process. One SQLite database. No third-party Python packages. Keep project state on infrastructure you control.
Inspect the source ↗One workspace. Many coding agents.
Run Attacca once, connect your tools, and give each project a durable place to coordinate.
Running in minutes.
Clone Attacca, start the local server, install the coding-tool integration, then connect your project.
# 1. Clone and start Attacca
git clone https://github.com/jdevgy/attacca-os.git
cd attacca-os
python3 attacca.py serve --host 127.0.0.1 --port 4173
# 2. In a second terminal, install integrations
python3 install.py --url http://127.0.0.1:4173
# 3. In your project, connect the coding agent
/attacca:setup
Read it. Run it. Improve it.
Attacca is developed in public on GitHub. Inspect the implementation, open an issue, contribute a change, or run the whole stack on your own machine.
Questions before you wire it into your stack?
What exactly does Attacca store?
Project memory, activity history, handoffs, tasks, claims, decisions, messages and other structured coordination state used by humans and connected coding agents.
Does Attacca replace Claude Code, Codex or Cursor?
No. Attacca sits beside coding tools as a shared project continuity layer. The coding agent still writes and reasons about code; Attacca preserves and coordinates the project state around that work.
Can multiple AI tools use the same project?
Yes. Attacca is designed so different AI coding tools and identities can connect to the same workspace while keeping identity, role, runtime and authority distinct.
Do I need a hosted Attacca account?
No. The project can run locally or on infrastructure you control. The core runtime uses Python’s standard library and SQLite.