A foundation, built deliberately.

A durable place to begin and return.

The following foundations are implemented and checked in the local development build. This is a partial foundation, not a general public release.

• Local work threads

  Capture an idea, append to its history, and read it again after restarting. The command-line interface and a read-only thread view use the same local record.

• Versioned working briefs

  Keep an objective, constraints, and a current summary linked to retained sources. Earlier versions remain available when the brief changes.

• Source-linked context receipts

  Record which thread sources were selected for context, why material was included or omitted, and whether the record is still current.

The complete session lifecycle, provider execution, and end-to-end approval and verification workflow are not yet available.

Make continuity more complete.

The next foundation work covers conversation sessions, deliberate forks and undo, and the remaining local lifecycle, export, retention, and deletion behavior. These need clear recovery paths before the product takes on more execution responsibility.

Carry the same clarity into execution.

• Explicit execution limits

  Bind work to an agreed scope, approvals, and a budget. Surface requests for consequential actions to the operator.

• Connections to AI tools

  Pass bounded work and context to supported tools, then collect the relevant results and execution records. No provider dispatch is active in the current foundation.

• Evidence-backed completion

  Connect a completion claim to the work, verification, and source records that support it. Keep unavailable evidence and unsupported controls visible.

• Model evaluation on your work

  Compare qualified models on a bounded, approved sample with results and cost coverage made explicit. Replay and automatic learning are not shipped capabilities.

Each layer has to show practical value on real work before it earns a place in the product.

Help shape the first useful release.

We want to hear from independent technical operators and founders who already switch between AI tools, return to long-running projects, and review work before it affects the outside world.

Tell us what you are doing and where the process loses its thread.

Request early access