Why White Tree

Same models. Different memory.

Soon every company will run on the same frontier models. The intelligence will be identical; the memory will not. White Tree exists so that the judgement your organization pays for every day becomes the asset that compounds, instead of the cost you pay again tomorrow.

The gap

Your systems store the work. The judgement evaporates.

Built for static information

Fifty years of enterprise software answered "where do we put the artefacts?" Documents, databases, files, search. Retrieval bolted on top rebuilds context from fragments, every single time.

Broken by the swarm

Thousands of agents querying in parallel pay the full price of reconstruction on every call. The archive grows as the company ages, so it gets more expensive exactly as you get wiser.

Knowledge debt compounds

Every lesson is paid for once: salaries, failed experiments, tokens. Without a memory it is paid again tomorrow, and an agentic company pays at machine speed.

The problem, in full →

The alternatives

Everyone is building agent memory. Nobody owns company memory.

Agent-memory tools

Zep, Mem0 and peers give a single agent a longer memory. Useful, and structurally per-agent: no organizational record, no governance, no authority model. Your company's judgement stays fragmented across a thousand private memories.

RAG and vectors

Retrieval over artefact storage answers "what is stored". It cannot represent which alternative lost, who authorised what, or that a belief was superseded last Tuesday, and it reprocesses the corpus on every query.

Build it yourself

Wiring memory into one workflow at a time, without bitemporal history, revocation or receipts: the parts of the problem that take years. Those parts are our product.

The difference

A trajectory database answers a different question.

Not "what is stored?" but "what happened, who learned it, and who may inherit it?" That question cannot be retrofitted onto artefact-shaped storage, so we made its requirements the engine's native primitives.

trajectories as records append-only bitemporal governed recall receipts from the engine model-agnostic

How the platform works →  The developer preview →

Principles

Constraints we will not trade away.

RECORD

History is never rewritten

Bitemporal, append-only. What was believed, and when, is preserved forever. Corrections add; they never erase.

TRUTH

Causation is never invented

The record links what the evidence links. Where causation is uncertain, it says so.

OWNERSHIP

You control the memory

Your memory is your asset. Never our training data, and it moves with you, across vendors and models.

PROOF

Every claim has a receipt

Access, recall and learning flows are provable after the fact, from the storage layer itself.

One company compounds every day. The other starts from zero, every morning. Alone.

See it in three and a half minutes.

The video makes the case better than any page.

Watch the video  Talk to the team