Caura

Caura is a governed memory service for fleets of AI agents. It gives every authorised agent in a tenant one persistent, shared memory, so knowledge learned by one agent is available to the agents that follow, while enforcing scope, policy, isolation and audit on every read, write and delete. The product was previously called MemClaw. It is built for platform teams deploying many agents inside an enterprise, and for the security, audit and finance stakeholders who have to sign off on them. Memory is LLM agnostic, harness agnostic and provider agnostic, and agents reach it through MCP, REST and Rail hooks. Governance is organised into four boundaries: organisation, tenant as the hard isolation line, fleet as the default collaboration boundary, and individual agent identity and attribution. The Caura Prism console maps what the fleet knows, replays and explains any recall, surfaces contradictions and stale facts, flags credential or personal data findings tied to the memory and agent that wrote them, and shows token cost per memory and per agent. Recall is tuned by outcomes, and repeated facts consolidate into knowledge. Pricing is published, with unlimited users, tenants, fleets and agents on every plan. Free is $0 with 10,000 memories, 5,000 writes, 5,000 searches and 500 recalls a month. Pro is $49 per month, or $490 a year, with 250,000 memories and higher limits. Business is $399 per month, or $3,990 a year, with 1,000,000 memories, custom models and a support SLA. Enterprise is quoted and adds on premises, air gapped or white label deployment, usually starting with a two week pilot. The engine is also open source under Apache 2.0 for self hosting, and the managed cloud states SOC 2 Type II compliance.