Open-core. pip install, self-host, and give your agents durable memory and an audit trail.
Open engine, MCP tools, clean Python API.
Sharded knowledge, FTS + vector, Context Broker.
Automatic PII cipher in/out of cloud models.
Orchestrate agents over a mesh with gated tool calls.
The technical capabilities behind this — from the live feature inventory.
pip install uaml-memory — OSS engine, MCP tools, a clean Python API. Studio is always free.
16 live shards (141k rows) + a team shard + fan-out search; Neo4j graph + 4096-dim embeddings + HNSW semantic search (45.5k gold nodes, similarity/recency/importance blend).
Point-in-time queries; memory_latest_for_topic walks the chain to the current canonical answer — vector stores don't do this.
Full producer & consumer of Google's OKF v0.1 (okf_import / okf_export), lossless round-trip even with Google's sample bundles.
Give each spawned subagent its own model profile — cheap/fast for mechanical steps, top model for the hard ones — mixing local Ollama and cloud per task; one router (OpenRouter + Ollama + native SDKs) with automatic fallback, bring-your-own-key per provider.
Free engine on your own hardware. No credit card. Founder pricing through 31 October 2026.
View pricing Back to home