# arca — memory for your coding agents

Your coding agents forget everything. arca doesn't.

arca indexes every Claude Code, Codex, and Cursor session on your machine
into structured, searchable memory. Any agent can search past decisions,
debugging trails, and teammates' reasoning — in one command.

## Install

```bash
curl -fsSL https://tryarca.ai/install | bash
```

One command: CLI, background daemon, and agent integrations (skills, hooks,
MCP) for Claude Code, Codex, and Cursor. Existing history indexes on the spot.

## How it works

1. **Install** — the installer sets up the CLI, a background daemon, and
   native skills + hooks for all three agents. Your existing history is
   indexed immediately.
2. **It indexes itself** — the daemon watches the agents' local session
   stores and indexes new sessions as they land: anchored sections (stable,
   citable excerpts), incremental and append-aware. No index command to run.
3. **Ask anywhere** — `arca context "oauth regression"` combines SQLite FTS,
   warm semantic recall, and mandatory reranking. Returns the three sections
   that matter, not a transcript dump. Typically < 50ms — and agents pull
   context themselves via hooks.

## Features

- Anchored sections, not transcript dumps (export exists when needed)
- Local-first: index, corpus, and daemon live in ~/.arca
- Strict filters: provider, time range, repo scope, cwd, subagent policy
- Warm daemon: no cold starts on the retrieval path
- Team memory: share sections with attribution (paid)
- Agent-native: skills for Claude Code / Codex / Cursor + MCP server

## For agents

Machine-readable docs: https://tryarca.ai/llms.txt
Setup skill: https://tryarca.ai/skill.md
Any page as markdown: append .md to the URL.

## Links

- Pricing: https://tryarca.ai/pricing.md
- GitHub: https://github.com/arca-sh/arca