The full open-source stack, hosted for teams: Mimir memory, Perseus live context, and Plutus spend monitoring — with shared vaults, team dashboards, SSO, and priority support. Context is injected before the model's window opens, in under 50ms.
Persistent memory — 36 MCP tools, AES-256-GCM encryption, FTS5 + vector hybrid recall, confidence decay. Shared team vaults in Cloud.
Live context injection — 26 MCP tools, 1,190× cache-punch throughput, 94% token compression, tiered disclosure (1–3).
Multi-provider spend monitoring — live balances, burn rate, and depletion forecasts across Anthropic, OpenAI, Google, Bedrock & Azure.
The engine is MIT-licensed and free forever, self-hosted. Cloud adds managed hosting and team features when you're ready.