Portfolio
Cawsal
A supervisor for agentic software engineering — governing what an autonomous agent changes in a live codebase, with human attestation where it matters.
Trust-boundPatent CandidatePrototypePatent-sensitive
01Category
Local-first / Trust-bound System
02Arc Role
Product Candidate / Patent Candidate
03R&D Funnel
Prototype
04Disclosure
Patent-sensitive
01
What it is
Cawsal supervises the work of coding agents. As autonomous systems begin to write and change software directly, Cawsal sits between the agent and the codebase: it reads the repository as a live system, examines what the agent proposes to change, and holds the boundary on what may proceed automatically and what a human must attest to.
02
Problem space
An agent can write code that runs and is still wrong — a silent deletion, a change that misreads intent, a modification with no test and no review. As more of a codebase comes to be written by agents, the question stops being whether it can write code and becomes whether what it wrote can be trusted to merge.
03
Arc's position
Arc builds the governance layer that lets agentic software engineering be trusted in production. The mechanism is protected; the discipline is public — that autonomous change to a system of record needs a supervisor, not just a prompt, and that uncertainty should be surfaced for human judgement rather than hidden. Cawsal is trust-bound by design.
04
Current status
Alpha
Related