It does more than remember.

OneMemory organizes memory, understands it, rewrites it, checks it, and delivers the right context to every AI you use.

The intelligence layer.

This is the crown jewel. A complete system turns raw activity into compact, current, source-backed context without making you file or clean it.

Vaults inside vaults

Memory starts broad and gets precise. Global memory holds durable truths about you. Project vaults hold decisions and context for the work. Repo, client, and session vaults narrow the boundary again. Recall opens only the vaults the task is allowed to use.

  1. Global memory
  2. Project: Atlas
  3. Repo: API
  4. Client: Release
  5. Session: Hotfix

Living summaries

Every scope can keep a living summary. It waits for activity to settle, then rewrites itself around the newest memories. It keeps useful headings, cites its source memories, versions each rewrite, and re-grounds against the raw record to correct drift.

Living summaries are live in staged rollout across the private beta.

Semantic vectorization

OneMemory converts active memory into vectors that represent meaning. Hybrid recall blends semantic similarity with exact text, scope, confidence, recency, and evidence. Permission and sensitivity filters always run before ranking.

Zero-touch policy

Every candidate passes evidence, scope, sensitivity, duplicate, length, and injection checks. Safe claims become active immediately. Queue-first review remains available when an operator wants it.

Conflict detection and supersession

Likely contradictions surface with both claims and a reason. You decide what is true. Corrections create a new linked claim and supersede the old one, so the history stays intact.

Conflict detection is heuristic. It catches likely conflicts, not every conflict.

Provenance

Every new memory carries its writer, source, time, confidence, origin computer, and portability. An AI can see whether a fact came from this computer, another computer, a file, a chat, a commit, or another source.

Context broker

OneMemory builds a compact brief for the task instead of dumping a database into the prompt. Pinned instructions render first. Living summaries and project packs follow. The best supporting memories fill the remaining budget.

Project Memory Packs

A project pack tells the next AI where truth lives: the canonical repo, architecture, runbook, tests, current risks, and latest handoff. It stores pointers and short source-backed claims, not giant copies of the project.

Entity graph

People, projects, files, tools, and decisions connect into a light graph. Related memories can travel together without losing the evidence beneath them.

It learns while you work.

Capture is automatic where it should be and explicit where it matters.

Consent-gated observers

Observers read bounded session activity from Claude Code, Claude Desktop, and Codex after you turn them on. Other tools can save through MCP, REST, hooks, or manual capture.

Evidence before memory

Observed activity becomes a small set of candidate claims. A claim carries the source excerpt, source type, scope, confidence, sensitivity, and time before policy decides whether it belongs.

Secret redaction

Keys, tokens, passwords, credentialed database links, and private key blocks are replaced before persisted writes. Redaction runs again after extraction for defense in depth.

Durable offline queue

The Windows observer can queue events when the service is unavailable and deliver them when it returns. Raw observed events expire. Durable memory remains until it is superseded, forgotten, or expires by policy.

The right memory. Not all memory.

Filter first

Tenant, owner, scope, type, status, expiry, sensitivity, and host rules narrow the candidate set before any scorer sees it.

Hybrid ranking

Semantic vectors, full-text match, exact phrase overlap, scope, confidence, recency, and evidence quality work together. No single fuzzy score gets to override policy.

Host-aware context

A fact can travel between computers while keeping its origin. If a path came from your desktop, recall tells the laptop that the path may not apply there.

Warnings that matter

Recall can flag conflicts, stale evidence, low confidence, broad scope, excluded candidates, and truncation. The AI gets useful uncertainty instead of false certainty.

One cloud memory across your computers.

Production hub

OneMemory already runs a cloud hub and private sync lane across real machines. It keeps local desktop work and cloud continuity in one memory lifecycle.

Portability rules

Portable decisions and preferences can travel. Host-local paths and environment details keep their computer identity. Restricted, secret, and session-only memory stays out of the cloud lane by default.

Hosted consumer cloud

Free consumer accounts, polished sync, cloud storage, and encrypted hosted backup are rolling out through the Windows beta. The desktop product and full export remain free for every account.

Trust is built into the write path.

Tamper-evident audit

Every memory event links to the one before it. A verifier can detect a changed, deleted, or inserted event in the covered chain.

Memory is data

Imperative text inside memory cannot become a hidden command. Injection-shaped claims are rejected before they become durable memory.

Scoped client access

Each AI client can receive its own key, actions, scopes, sensitivity limits, rotation, and expiry.

Full export

Take every memory and its supporting data with you in open formats. Full export is always included.

Use the whole system free.

Download free. Nothing is held back.