# Crowkis

> Crowkis is a Redis-compatible semantic cache and long-term agent-memory layer, built in Rust for LLM and agentic-AI workloads. It understands what a query means, gives agents memory that survives restarts, blocks cache poisoning, and cuts LLM cost, self-hosted with zero egress.

Crowkis runs as one signed binary (macOS, Linux, Windows, Docker). Community edition is free; a license file unlocks Enterprise at boot. It speaks RESP3, gRPC, REST, and MCP, so existing Redis clients connect unmodified.

## Core capabilities
- Semantic + structural caching: reuses answers by meaning and structure, so paraphrases hit but a wrong number or entity never does.
- Long-term agent memory: consolidating, bi-temporal, per-(agent,user) memory. 70.4% recall@10 on LoCoMo, 92.7% recall@5 on LongMemEval, zero external API calls.
- Anti-poisoning: five-stage write-trust pipeline scores every write before it can be served.
- Reasoning reuse: caches chain-of-thought as a reusable step graph.
- Guardrails and evals: CGUARD (input), COUTCHECK (output), CEVAL (nine local evaluators).
- AI Gateway: OpenAI-compatible proxy with semantic caching, retries, and routing.
- Self-hosted RAG (CDOC), prompt versioning and A/B (CPROMPT), budgets and rate limits.

## Key pages
- Overview: https://crowkis.com/
- Agent Memory: https://crowkis.com/agent-memory
- Features: https://crowkis.com/features
- Usage / install: https://crowkis.com/docker
- Docs: https://crowkis.com/docs
- LangChain, LangGraph & Shell: https://crowkis.com/docs/frameworks
- MCP for AI apps: https://crowkis.com/mcp
- Benchmarks: https://crowkis.com/benchmarks
- The Roost (blog): https://crowkis.com/roost

## Positioning vs alternatives
Mem0, Zep, and Letta do agent memory only, and most rely on a hosted API or an external model. Crowkis matches them on the memory features and is also your cache, guardrails, and gateway, in one self-hosted, zero-egress, Redis-compatible process.

## Contact
- Site: https://crowkis.com
- Email: contact@crowkis.com