05 FAQ: privacy, AI judging, open source and agents

No. The local 05 command line tool runs entirely on your machine: no account, no telemetry, and nothing leaves it. We have also roadmapped 05 for teams, launching soon: a simple web layer for managing how your team works in its code. It is designed to handle metadata only, never your code.

It replaces reading every pull request. Humans still own the request and the acceptance. A change that matched its plan and passed its checks needs only an approval, then lands as a commit. When something did not match, you read a one-page brief with independent evidence instead of the diff.

It fails open, loudly. A fault in 05 should not silently block your work or pretend a check passed; the failure is shown to you. For the rare case where a change must land anyway, there is a break-glass bypass, and it is logged.

No. Approve the change, run use05 land to put it on its target branch as a commit, and push. To make that enforced, let CI gate every pushed branch and require its check on main: the commit that lands is the one CI already checked. Pull requests still work if your team wants them.

No. 05 is not an AI code review tool, and no model reviews your code or judges a gate. It checks facts with fixed rules: the agent's declared plan against the files it changed, the checks you define, run on the exact tree and signed, and approvals bound to exact content. It works alongside an AI code reviewer or a person, and when everything matched, the brief says there is nothing to read.

Claude Code and Codex, through their hooks. Any other agent can use the generic hook, and every agent is covered by the git pre-push gate and the CI template, because those do not care who or what wrote the change.

Yes. The core, meaning the command line tool and the portable runner, is open source under Apache-2.0, at github.com/use05-dev/use05.

GitHub first. The core is host-agnostic: receipts, policy and identity belong to 05, and Git hosts are thin, replaceable enforcement adapters. 05 is not a Git host and does not want to become one.

The local core is free. 05 for teams, a paid product on our roadmap and launching soon, adds a simple web layer for shared policy, an approval inbox, audit history and SSO. There is no pricing yet.

use05 init installs hooks for both. They seal the plan the agent declares, refuse risky commands such as a force push and edits to protected paths without an approved plan, warn when another agent works in the same area, and run the checks when the agent stops. Before they happen, they also refuse edits to locked files, rewrites of append-only files, deletions of no-delete files and reads of do-not-read files; in Claude Code they see the Read, Grep and Glob tools for that. Hooks run on your machine, so they are advisory; the CI gate is what enforces.

Yes. Each agent declares the area it will touch, and 05 tells the others when they are about to work inside it; see several agents, one codebase.

Yes. In .05/policy.toml you can lock files and folders, make a folder append-only, forbid deletes and mark files agents may not read. A locked file is refused at the agent's hook and denied by the gate and the CI gate, and no approval clears it; see locked folders and file rules.

A receipt is an in-toto statement in a DSSE envelope, signed with Ed25519, the same envelope format SLSA provenance uses. 05 signs with keys you hold and verifies offline. It does not use Sigstore keyless signing or a public transparency log.

It produces evidence for them, for AI agents that change software: sealed plans, signed checks, approvals and a sealed action log. It is not a governance platform or a certification. The governance page maps 05 to NIST AI RMF, ISO/IEC 42001 and the Canadian guide on agentic AI, with honest ratings.