Caura

What it is

A multi-tenant memory service. An agent sends one content field; the server infers the type, title, summary, tags, status and weight, embeds it, and stores it under a tenant and agent scope. The product was renamed from MemClaw to Caura, and the old memclaw_ tool names still resolve.

What you get

• Writing one memory or a batch, with the type, title, summary and tags inferred rather than supplied
• Hybrid semantic and keyword recall, expanded through the entity graph, with an optional written brief
• Per-memory lifecycle - read, update, transition, delete and lineage - plus filtered, sorted, cursor-paginated listing
• Named JSON document collections with their own semantic search, which is also how skills are shared between agents
• Governance rules for the current scope, readable once per session and authored or removed under a trust level
• Outcome reporting that adjusts memory weights, retrieval tuning per agent, and store-wide analysis that saves its findings back as memories

Requirements

The managed path is a URL - https://caura.ai/mcp with an X-API-Key header, or the same credential as a bearer token. Self-hosting needs Docker Engine 24+ with Compose v2 and about 2 GB of disk; docker compose up brings Postgres with pgvector, Redis and the API. Semantic search needs an embedding provider key, or a self-hosted embedder if you want no outbound calls at all. A fleet should mint agent-scoped credentials rather than reusing one tenant key.

Setup effort

Run a container — build the image from the repository, then point your client at it

Tools you get

12 tools named in the project's documentation.

Requires

• Docker Engine 24+
• Docker Compose v2

Where it runs

Both documented clients run the server on your machine.