Questions, answered plainly.

You assign a GitLab issue to its bot. It implements the issue, runs your tests, waits for your pipeline, scans the change and has it reviewed for security problems, fixing what it finds. Then it hands you a merge request that is ready to approve.

No. It pushes to its own branch and opens the merge request. Merging stays with your team and your branch protection.

Ones a colleague could pick up from what is written down: bug fixes, small features, refactors, missing tests. For a large or open question, label it nightsquire::plan and get a plan first.

Yes. Comment on its merge request and it makes the change, runs every check again and answers in the thread.

It continues once on its own if it was making progress. If it still can't finish, it commits the work so far, leaves the merge request as a draft and comments where the work is and how to continue. You get an email either way.

No. Nightsquire is built for GitLab, on gitlab.com and self-managed. If you know the coding agents on GitHub, here is how Nightsquire compares.

No. Any tier works, Free included, on gitlab.com or your own instance. On gitlab.com Free you give it a bot account of your own, because group access tokens need Premium there.

Nothing, if you pick a managed executor in one of three sizes. Your own needs a Linux VM with systemd on amd64 (x86-64): Debian, Ubuntu and their derivatives, RHEL, Rocky Linux, AlmaLinux, Oracle Linux, CentOS Stream or Fedora. It needs outbound HTTPS only, no inbound ports. Size it like a CI runner for the projects it serves.

With one command, which the dashboard gives you together with a registration token. It installs Docker Engine from docker.com, replacing an older or distribution-packaged Docker, or Podman's Docker emulation, so use a VM dedicated to Nightsquire. The installer is the only way to install one.

Nightsquire AI, which works without any setup and runs in the EU. Or your own Claude API key, or any OpenAI-compatible API: OpenAI, Mistral, OpenRouter, Azure OpenAI, or your own model on vLLM or Ollama. Pick a model with reliable tool calling; small local models struggle with real issues.

Yes. It reads each project's CLAUDE.md, and can suggest one for you. Notes about your GitLab instance, written once in the dashboard, go to every run.

Yes, per project. The step gets its own Docker daemon in a sandboxed sidecar, never your host's socket, with a cache that survives between runs.

No. Steps get placeholders, and the gateway on your host swaps them for the real credential on the way out. How that works

It is checked out and run on your executors, managed or your own. Run state passed between steps, logs and transcripts can contain code and are stored by Nightsquire for the retention period you set, 90 days by default.

Members of your organisation with the Member role or higher, recognised by their GitLab account. You can narrow it per project.

Yes. Every host a step reaches is recorded. Switch a project to an allowlist built from that list whenever you are ready.

Only in the EU: the control panel on netcup in Germany, and managed executors, Nightsquire AI and email on Scaleway in France. Nightsquire is built and run by an Irish business, under Irish law, and those two are its only sub-processors. Our data processing agreement is part of the terms. Data sovereignty

Nothing Nightsquire runs, Nightsquire AI and managed executors included. Only a GitLab or model provider of your own outside the EU takes anything out, under its own terms.

Almost. With self-managed GitLab, your own executors and a model on vLLM or Ollama, no model provider sees your code. Run state, logs and transcripts still go to Nightsquire, in the EU, for the retention period you set, and your GitLab must be reachable from Nightsquire over HTTPS.