Docs
Everything to evaluate, install, and operate Heartwood Memory. A full hosted docs portal is on the way — for now, deep references live in the repository.
Quickstart
Zero to a governed recall — embedded Python, HTTP, and MCP.
Integration guides
Markdown import, warm recall, MCP server, bulk API, and end-to-end onboarding.
Concepts
Typed memory, provenance, policy-before-ranking, faithfulness, deletion lineage.
Python & TypeScript API
Full reference for remember, recall, explain_recall, approve, forget.
Deployment
Docker runtime, MCP server, and embedded usage beside your stack.
Governance & trust suite
The executable gates behind every claim — run them yourself.
Key custody
Envelope encryption, crypto-shred DEKs, and KMS/HSM-compatible custody.
Install and run
Pick a surface and go. The core is source-available and self-hosted — your data never leaves your environment.
On 0.1.0? Upgrade to 0.1.1 — it fixes a memory leak in the warm recall service (serve-recall) that could grow resident memory unbounded under sustained load. pip install -U "heartwood-memory[recall,mcp]"
The embedded library. Remember and recall in a few lines, governed by default.
python -m pip install "heartwood-memory[recall,mcp]"from heartwood import Heartwood, Principal
hw = Heartwood(path="./heartwood.db", tenant="tenant:acme")
hw.remember(
"Refunds over $500 require finance approval.",
subject="policy:refunds",
created_by="agent:support",
)
principal = Principal(
id="agent:support",
tenant="tenant:acme",
clearance="internal",
)
results = hw.recall("what is our refund policy?",
principal=principal)