# cachly — AI Brain for coding assistants

> cachly gives AI coding assistants (Claude Code, Cursor, GitHub Copilot, Windsurf, Cline, Zed) a persistent memory that survives across sessions. The Brain stores lessons, past fixes, architecture decisions, and session context — and recalls them in under 10 ms.

## Product

cachly is an MCP (Model Context Protocol) server + managed Brain backend. When installed, the AI coding assistant automatically:
- Starts a session on editor connect (no manual call needed as of v0.10.9)
- Loads the previous session summary, open tasks, and top lessons
- Stores what it learns via `learn_from_attempts()` after every fix or deploy
- Recalls past solutions via `recall_best_solution()` and `smart_recall()`

**Key facts:**
- 126 MCP tools (session management, memory, search, causal graph, prediction)
- Sessions fire automatically on `ListTools` (editor connection)
- Lessons persist across machines, editors, and engineers
- Causal Knowledge Graph with Bayesian confidence scores
- VS Code extension with offline queue (lessons saved without network)
- Free tier, EU servers (Germany), GDPR-native
- npm: `@cachly-dev/mcp-server` — 11,953 downloads in first 23 days

## Quick start

```bash
npx @cachly-dev/mcp-server@latest autopilot
```

Or add to `.mcp.json`:
```json
{
  "mcpServers": {
    "cachly": {
      "command": "npx",
      "args": ["-y", "@cachly-dev/mcp-server@latest"],
      "env": {
        "CACHLY_JWT": "your-api-key",
        "CACHLY_BRAIN_INSTANCE_ID": "your-instance-uuid"
      }
    }
  }
}
```

## Core MCP tools

- `session_start` — 🤖 Auto (fires on editor connect). Returns previous session summary, open tasks, top lessons.
- `session_end` — 🤖 Auto (fires on editor disconnect). Generates session summary from git log.
- `learn_from_attempts` — ⭐ Most important. Store what worked/failed after every fix or deploy.
- `recall_best_solution` — ⭐ Before any non-trivial task. Returns best known approach with confidence.
- `smart_recall` — Semantic search over all stored lessons by meaning.
- `remember_context` — Store arbitrary key-value state (file summaries, config, progress).
- `recall_context` — Retrieve stored context. Supports glob patterns (`file:*`).
- `session_handoff` — Store open tasks before switching editors or windows.
- `brain_predict` — Predict likely failures for a given context. Pre-load mitigations.
- `ckg_inspect` — Query the Causal Knowledge Graph directly (typed edges, confidence scores).
- `index_project` — 🤖 Auto (once per 24h). Index the codebase for context-aware session starts.
- `brain_search` — BM25+ full-text search over all brain data.

## Supported editors

- Claude Code (Anthropic)
- Cursor
- GitHub Copilot (VS Code)
- Windsurf (Codeium)
- Cline
- Zed
- Any MCP-compatible editor

## Key concept: learn → recall loop

```
1. AI fixes a bug → calls learn_from_attempts(topic="fix:...", outcome="success", what_worked="...")
2. Next session → session_start returns top lessons automatically
3. Before similar task → recall_best_solution("fix:...") returns fix in <10ms
4. Brain predicts failures → brain_predict(context="deploying X") returns mitigations
```

## Pricing

- Free: 1 Brain instance, 500 lessons, 1,000 recalls/day
- Speed (€79/mo): 3 instances, 5,000 lessons, unlimited recalls
- Business (€199/mo): 10 instances, unlimited lessons, team sharing, priority support

## Links

- Homepage: https://cachly.dev
- Sign up (free): https://cachly.dev/sign-up
- MCP docs: https://cachly.dev/docs/mcp
- Recall benchmark (methodology + results): https://cachly.dev/docs/benchmark
- Agents docs: https://cachly.dev/docs/agents
- IDE plugins: https://cachly.dev/docs/ide-plugins
- Blog: https://cachly.dev/blog
- npm: https://www.npmjs.com/package/@cachly-dev/mcp-server
- GitHub: https://github.com/cachly-dev/cachly
- API docs: https://cachly.dev/docs/api

## About

cachly is built in Germany. Servers in Nuremberg, Germany (EU). GDPR-native — no tracking cookies, no PII in memory, full data portability (GDPR Art. 20), right to erasure (GDPR Art. 17).