Why Signoos is built the way it is
AI can now read a codebase, review a change and draft a patch. The question that decides whether any of that helps a team is older than AI: who decided, on what evidence, and where is the record? Signoos is built around that question.
Delegation needs a record
When work is delegated in a chat message, the ask, the code it concerns, the review and the decision scatter across four tools. Signoos keeps them as one object: a task holds the ask in the lead's own words, the files the localizer bound it to, the branch the work is judged on, every advisory review, and the conversation — including disagreement. Six months later, “why is the code like this?” has an answer you can open.
Decisions stay human
Every AI feature in Signoos produces advice or a draft — never an action on your repository. Reviews cannot block a merge. Verdicts cannot close a task. The solver's output is a patch on a thread that a person applies, or does not. This is not a limitation we accepted; it is the design. A tool that acts on your repository has to be watched. A tool that advises can be trusted with more, because the blast radius of a wrong answer is a conversation, not an incident.
Evidence over assertion
Everything Signoos computes stays attached to what it was computed from. A push review cites the ask it measured the push against. An Insights gap resolves to the verbatim quotes behind it and the roadmap items it was tested against, with confidence computed rather than asserted. When the AI cannot find something — a file, a supporting signal — it says so and shows its candidates instead of guessing. A dashboard you cannot interrogate is a rumour with a user interface.
Consent you can point to
Most tools ask for repository access once, at install, in full. Signoos asks a better question: which folders, for which project, revocable at any time. The grant is enforced in the one place every read goes through — the index, the localizer, the mirror and the file browser alike — and the check fails closed. Your security review does not have to take our word for it: the security overview describes the mechanism, and the AI processing page lists what leaves, job by job, with the caps.
Nothing to leak
Signoos stores no GitHub token, no password for your repository, no long-lived key to your code. Each App installation mints an hour-long token at the moment of use and lets it expire. The solver's sandbox runs with no credentials at all. Provider API keys a team chooses to bring are sealed with envelope encryption and are never returned by any endpoint. The cheapest breach to survive is the one where there was nothing to take.
Model choice without lock-in
Models improve monthly, and the best one for reviewing a push is not the best one for titling a task. In Signoos each AI job is a role, and the team leader assigns a model to each from a catalogue across OpenAI, Anthropic and Google — per project, changeable at any time, visible to every member. Daily spend ceilings make experimentation safe; bring-your-own keys make the bill yours when you want it to be. You are never married to a provider because your tool was.
A product that says so
A role running on a fallback model says so instead of pretending. A diff with files withheld by consent tells the model — and you — how many. A review that finds nothing wrong says that, and an Insights pass with no supporting signal reports it as a result, not an error. This page has no invented metrics on it for the same reason. Teams run on trust; tools should be built the way teams run.
Where to start
Start with one repository: connect it, answer one question about code access, and delegate a task. The documentation walks through the first ten minutes, including the terminal client and the editor integration. You can change every answer, or disconnect entirely, whenever you like.