The questions engineers ask first.

No. It creates a branch named task/<id>, commits the suggestion you chose, and opens a pull request. Merging stays with your review rules and your CI.

Tracker credentials (Jira, Azure DevOps) and repository credentials (GitHub, Azure Repos) are stored against your user, tested when you save them, and never written to a log line. Each one is encrypted with AES-256-GCM before it is written, under a key held in the environment rather than in the database, so a copy of the database on its own does not yield a usable credential. Model credentials are yours too: Kandryn calls Anthropic and OpenAI with the keys you save, so the usage appears on your own account and no key is shared between users.

A handful of files — the ones the change planner selects as relevant to the work item — plus the detected stack profile. The planner is also shown the directory listing, so it can target paths that exist, but that is file names only and is capped. No other file contents leave the repository, and nothing outside the repository you bound to the project is read at all.

The run row records the error, the item is left untouched, and the owner gets an email. Runs stuck longer than twenty minutes are swept to failed by the dispatcher.

Only where you ask it to: new items and test cases you explicitly push, and a status change when an item closes. Nothing else propagates upstream.

GitHub is the primary, auto-synced provider. Azure Repos handles branches, commits to existing files, pull requests, stack detection and the security gate; adding a brand-new file can fail there, and the gate needs a pull request to attach to, where GitHub can post against the commit alone.

Each agent has a distinct role. Raptia and Fovea generate competing suggestions in parallel — they reason differently by design, so when one misreads the ticket, the other usually does not. Synthesia scores both and recommends the stronger answer out of 10. Veria reviews the committed code against the acceptance criteria after commit. Aegis scans for security vulnerabilities and fails its status check on a High finding, which blocks the merge once your branch rules require that check. Narratia writes the operational runbook. Together they cover the full delivery loop from generation to documentation.

Aegis checks for OWASP Top 10 (2021) vulnerabilities: injection flaws (SQL, NoSQL, command), authentication and authorisation flaws, cryptographic weaknesses, hardcoded secrets, insecure data exposure, missing input validation, and SSRF, XXE and deserialization issues. It reads one changed file at a time with no cross-file or dependency-manifest context, so categories that depend on how files relate — broken access control, vulnerable components — are largely out of its reach. Each finding has a severity (Critical, High, Medium, Low, Info), an OWASP category, a line reference, and a remediation step. High and Critical findings fail the status check Aegis posts to the pull request — a block once your branch rules require it. Medium and Low findings create tracker tickets.

Yes — that is what Remediate Now is for. Click it on any Aegis finding and Kandryn creates the tracker ticket, syncs it to the board, and immediately starts a new run with the security finding and its remediation as the brief for Raptia and Fovea. The loop closes in the same session without switching tools.

Eight sections: a summary, what changed and why, deployment steps specific to this change, a rollback procedure referencing the branch and PR, validation commands to confirm it is working in production, the generated test cases, security findings from Aegis, and a references section with the work item key, branch, commit hash, and PR link. Pushed to Confluence, Notion, or committed as Markdown to the same PR branch.

That is how every pilot starts: one tracker project, one repository, one real work item run end to end on a shared call.