An AI coding agent for your own GitLab.
On your hosts, if you want.
Nightsquire works on self-managed GitLab on any tier, Free included. Assign an issue and get back a merge request that has been through your tests, your pipeline and a security review, with the agent's work running on executors you host and the model you choose.
Four steps on your own instance.
Your beta invite opens a setup wizard that tests each step before you move on.
- 1
Connect your instance
Create an OAuth application on your GitLab with the values the wizard gives you. Your team signs in with that GitLab, so its sign-in rules still apply.
- 2
Let it create the bot
Paste an admin token once and Nightsquire creates its own bot user. Its token rotates itself from then on.
- 3
Run an executor next to it
One command on a Linux VM in your own network or cloud. It opens no inbound ports and connects out over HTTPS.
- 4
Pick a model, enable projects
Nightsquire AI, your Claude API key, or a model you host. Enable a project and its webhook and bot membership are set up for you.
Keep the code, the commands and the model inside your network.
- Your code is checked out and run only on executors you host.
- Prompts go from your executor straight to a model on vLLM or Ollama, so no model provider sees your code.
- Every host a step reaches is recorded; lock a project to an allowlist built from what it used.
- Docker for tests runs in a sandboxed sidecar, never on the host's socket.
The rest stays in the EU
Nightsquire itself runs only on European clouds, netcup in Germany and Scaleway in France, from an Irish business, under Irish law and the GDPR.
- Settings and tokens, encrypted with your organisation's key
- Run state, logs and transcripts, for the retention you set
- The audit log of every sign-in, grant and config change
It fits the way you already run GitLab.
No licence change
Any tier of self-managed GitLab works, Free included. No Premium, no Ultimate, no Duo.
Your sign-in rules
Your team signs in with your own GitLab, so the SSO, two-factor and access rules you already enforce apply to Nightsquire too.
Least privilege in GitLab
The bot gets Developer on the projects you enable, pushes only to its own branches and never merges. A run reads only what the person who started it can read.
Credentials out of reach
Every step gets a placeholder. The gateway on your executor swaps it for the real token on the way out, so the agent's code never holds it.
Settings you can audit
Set limits, scanners and network rules for the organisation, override them per project, and see every change in a versioned, audited history.
Managed when you'd rather
Don't want to run a VM? Pick a managed executor instead, dedicated to you, on Scaleway in Paris.
What admins ask.
Do we need GitLab Premium, Ultimate or Duo?
No. Nightsquire works on any tier of self-managed GitLab, Free included, and doesn't use Duo.
Can we keep our code on our own hardware?
Yes. With your own executors and a model on vLLM or Ollama, your code is checked out, run and sent to the model only on your hosts. Run state, logs and transcripts are kept by Nightsquire, in the EU, for the retention period you set.
How does Nightsquire connect to our GitLab?
Over HTTPS: your GitLab sends webhooks to Nightsquire, and Nightsquire uses the GitLab API and your GitLab sign-in. Executors only connect out, so they need no inbound ports.
What does an executor need?
A Linux VM with systemd on amd64 (x86-64): Debian, Ubuntu, RHEL, Rocky Linux, AlmaLinux, Oracle Linux, CentOS Stream or Fedora. Outbound HTTPS only. Size it like a CI runner for the projects it serves.
What permissions does the bot get?
Developer on the projects you enable. It pushes only to its own branches, creates issues only when asked and never merges. Where a run needs another project, an Owner or Admin can grant it Reporter, never more.
Point it at your own GitLab.
Nightsquire is in a closed beta. Leave your email and we'll tell you when it opens to more teams.