cachly: memory for your AI coding assistant
npx @cachly-dev/mcp-server@latest autopilotOne command signs you in, detects every editor you use, writes the correct MCP config for each one, and bootstraps your Brain from your entire git history. Free forever, no credit card, EU servers.
Why teams add cachly#
- Zero re-explaining. Your AI opens each session already knowing your stack, your open bugs, and what the team learned yesterday.
- Automatic learning. A post-commit git hook feeds every fix, revert, and decision into the Brain — no manual note-taking.
- Team-wide, not per-user. One engineer's solved bug becomes the whole team's reflex, searchable in any language.
- Model-neutral. Switch from Claude Code to Cursor to Windsurf and the memory follows — your Brain lives in your own Valkey instance, not inside a model vendor.
Start here#
| Section | What you'll find |
|---|---|
| AI Memory | The 3-layer memory system: storage, tools, and the autopilot file that wires them together |
| MCP Integration | Full MCP server reference — automatic sessions, Ambient Recall hooks, supported editors |
| Memory Crystals | Compressing thousands of lessons into a dense per-session digest |
| Model-Neutral Brain | Why your memory isn't locked to one AI vendor, and how to verify it |
| Agents SDK | Give LangChain, CrewAI, AutoGen, and custom agent loops persistent memory |
| IDE Plugins | VS Code and JetBrains extensions that show what your AI has learned |
| Cache Engines | Dragonfly vs. Valkey — how to pick the right engine for your workload |
| Semantic Caching | Vector-similarity cache hits for repeated LLM queries |
| Cluster Mode | Horizontal sharding, automatic failover, smart routing |
| Terraform Provider | Manage cache instances and API keys as code |
| Team Knowledge Hub | Browse team lessons, prove retention after offboarding, generate onboarding briefs |
| API Reference | REST and gRPC surfaces, authentication schemes, endpoint map |
| Recall Benchmark | The measured, reproducible lift cachly's ranking gets over flat-file memory |
How the pieces fit together#
cachly is not a semantic cache with a memory feature bolted on — the Brain is the product, and the semantic cache is the proof-point that shows savings from day one. Under the hood, both run on Redis-protocol-compatible engines (Dragonfly or Valkey), while the memory layer itself is exposed as 122 MCP tools that any MCP-compatible client can call.
Read the AI Memory page first if you're setting cachly up for the first time — it covers the fastest path from zero to a briefed AI assistant.