← All projects

02 / Local intelligence

dawnr

READ · PLAN · CHECK

Your computer.
An assistant that stays local.

An offline assistant for documents, files, and computer tasks, with reviewable changes, sandboxed commands, and a path to formally checked code.

View repository
Offlineworks locally by default
17 / 182problems proved by at least one prover
12 / 182problems proved by all seven

From the public project record reviewed 08 October 2026. View source and measurement context ↗

Inside the project

Built around
the details.

01

Work with the files you already have

Read text, PDFs, office documents, saved pages, and mail. File changes are presented as plans, and applied changes are journaled so they can be reviewed and undone.

02

Keep control of the machine

Commands run in a sandbox. Actions on the computer are shown for approval. Network access is off by default, and the assistant does not require a cloud account.

03

Judge the outcome

The evaluation examines resulting files, commands, and machine state. For code, dawnr combines tests, specification checks, and the proof engine rather than relying on the model’s description of its answer.

What the results mean

dawnr is experimental. Its published evaluations include harmful mistakes and the checks added in response. Review actions before approving them. The 17 and 12 proof results refer to dawnr v5 on a clean set of 182 programming problems; they are not a general success rate for computer tasks.

Follow the evidence

Source & records.

Next in the collection / 03locallm