White Tree is organizational learning infrastructure for the agentic enterprise. Your company decides things all day; almost none of that judgement survives the week. We built the memory where it compounds instead.
The agentic future is here. It is being built on yesterday's infrastructure.
One 500-person company, one year from now. Every one of those minds starts from zero, every time. The full problem →
White Tree is the trajectory store and governance engine; Dúnedain is the capture and enforcement runtime. Every surface is append-only, authority-checked and receipt-emitting, and sits below the model layer, so nothing you learn is hostage to model churn.
Append-only, bitemporal storage where the unit is the decision thread, not the row. History cannot be silently rewritten, by anyone.
The record → harnessRides your chat, code, tickets and CRM, extracting decisions as a by-product of work. No forms, no knowledge-management chores.
The runtime → recallAuthority, expiry and revocation evaluated in the engine on every read. Every recall emits a verifiable receipt at the storage layer.
How authority works → mcpSix tools any agent can call: recall, search, trace, capture, act, grant. Claude today, whatever wins next year: the memory is untouched.
The tools → piiRedaction-aware extraction with a full audit trail. The lesson generalises; the sensitive substrate stays sealed to those cleared for it.
Privacy model → federationGoverned slices of memory, consultable across company lines and revocable the day the relationship changes. The network effect, engineered.
The research →Point the harness at the systems you already run, then give any MCP-capable agent the memory. Authority rides on the connection; receipts are emitted on every operation.
# the harness rides alongside, it does not replace sources: - slack: { channels: [deals-*, eng-*] } - github: { orgs: [acme-inc] } - jira: { projects: [PRICE, PLAT] } - salesforce: { objects: [Opportunity, Case] } extraction: mode: trajectories # decisions, not documents pii: strider # redaction-aware, audit-trailed authority: receipts: always
const thread = await wt.recall({ question: "Why did we choose usage-based pricing for APAC?", as: "agent:sdr-07", // authority checked on every read scope: "commercial/pricing", }); thread.decision; // the call, who made it, on what basis thread.alternatives; // what lost, and why thread.outcome; // what the world said back thread.receipt; // verifiable proof this recall happened
Per-company isolated deployment. Fastest start; your memory never trains anything and leaves with you if you go.
Engine and harness inside your cloud account. Your keys, your network boundary, our operations tooling.
Air-gapped options for regulated estates, Strider redaction at the boundary, receipts auditable on your side.
Claude
ChatGPT
Codex
Copilot
Gemini
Cursor
Slack
Jira
Salesforce
Workspace
Notion
ZendeskSoon, every company will run on the same models. The intelligence will be identical. The memory will not.
One company compounds every day. The other starts from zero, every morning. Alone.
Quickstart, the trajectory record, the MCP tools, deployment options.
Read → whyThe swarm, the archive, knowledge debt, and the amnesia montage.
Read → howTrajectories, governance, token economics, and the harness.
Read → whereThe larger agent, the network of learning organisms, and the fork.
Read → whoWhy "White Tree", why "Dúnedain", and the team.
Read →Same models. Different memory. One asset compounds.
Get in touch