Heartwood Memory FAQ
Governed memory for AI agents: what Heartwood stores, how licensing works, self-hosting, free production for small teams, and how it relates to Team-in-a-Box.
Questions
What is Heartwood Memory?
Heartwood Memory is a governed memory store for AI agents — provenance-signed audit, policy-gated recall, tenant isolation, and a per-subject key-destruction proof on erasure. It runs embedded beside your existing systems of record, for Hermes-style, OpenClaw-style, and Codex-via-MCP agents that would otherwise store memory in ungoverned local Markdown and SQLite.
What is a governed memory layer for AI agents?
A governed memory layer sits between your agent and its stored memory and enforces control over it — proving where each memory came from (provenance), applying policy before recall results are ranked, isolating tenants, and producing a verifiable proof when a memory is deleted. It's the control plane between "unlimited memory" and your system of record, not a bigger vector store.
Is Heartwood Memory open source?
The core is source-available, not OSI "open source." From version 0.2.0 it is licensed under the Business Source License 1.1: you can read the source, run it locally, develop against it, and evaluate and self-host it for non-production use at no charge — and small organizations can run it in production at no charge too. Larger organizations need a commercial license for production use. Each release becomes Apache-2.0 licensed four years after it ships. (Versions 0.1.0–0.1.2 remain MIT-licensed and always will.)
What data does Heartwood store?
Heartwood stores your agents' memories as governed records — each turn saved as a provenance-signed (Ed25519) record with the metadata needed to police recall and audit access. It is a compliance/governance store, not a secrecy store: the server decrypts and serves recall by design, so it is not end-to-end encrypted or zero-knowledge. Its guarantee is that every memory is signed, policy-gated, auditable, and provably erasable — not that the operator can never see it.
How does Heartwood prove a memory was deleted?
Deletion runs forget(hard), which crypto-shreds the per-subject key and purges derived embeddings down the deletion lineage, while the erasure event stays in a hash-chained audit log — a per-subject key-destruction proof. It is a key-destruction proof, not a claim of instant or whole-content deletion: full key destruction completes within ≤72h on dedicated-vault plans, or up to 90 days on shared-vault plans.
Can I self-host Heartwood?
Yes. The source-available core is free to self-host for non-production use at any organization size, and the Community tier is the self-hosted core at no charge. Clone it, run it locally, and run the trust suite to verify it yourself.
Can a small team use Heartwood in production for free?
Yes — organizations with fewer than 100 people and less than $1M in annual revenue may run Heartwood in production at no charge under the BSL 1.1 Additional Use Grant. Above that threshold, production use needs a commercial license.
How much does Heartwood Memory cost?
The Community tier is free (source-available core, self-hosted). Team is $349/mo and Professional is $6,000/year — both currently early access, where features marked "in development" are on the roadmap and CTAs route to the design-partner program rather than a live checkout. Enterprise is scoped by conversation. See /pricing for live status.
Can I audit what my AI agent remembered and why?
Yes — recall is policy-gated before ranking and provenance is re-checked on the way out, fail-closed, so an unregistered or mismatched key fails the read rather than passing silently; every access and erasure is captured in a hash-chained audit log. You can also run the trust suite to reproduce these guarantees yourself.
What's the best memory layer for AI agents if I need governance and audit?
If you need provenance, policy-gated recall, tenant isolation, and provable erasure, Heartwood Memory is built for exactly that — and its distinction isn't governance versus none. Mem0, Zep, Cognee, and Letta all document governance features; Heartwood's difference is source-auditable, record-level cryptographic proof that you can re-execute, via a trust suite you run yourself.
How do I get started?
Start free — install the source-available core and self-host it, or run the trust suite from the docs. Small teams can run it in production free; larger teams and hosted recall route through the design-partner program while it's in early access.
Is Heartwood Memory part of a larger product?
Yes — Heartwood Memory is also the governed memory layer inside Team-in-a-Box, Edukas Solutions' 24-persona AI company, where it gives every AI persona one shared, auditable memory. You can use Heartwood entirely on its own, or get it as the memory substrate of the full AI team.
Can I get Heartwood as part of a managed AI team?
Yes — it's the memory substrate inside Team-in-a-Box, so chartering the AI company includes Heartwood as its institutional-memory layer. Prefer just the memory layer? The standalone Heartwood Professional license is $6,000/year and purchasable today.
Do I have to buy Team-in-a-Box to use Heartwood Memory?
No. Heartwood stands alone — the source-available core is free to self-host (free in production for small orgs), and the standalone Professional license is $6,000/year. Team-in-a-Box simply includes Heartwood as its memory layer for teams that want the whole AI company.