Claude on your GitLab issues.
Without building the pipeline.
Give Nightsquire your Claude API key, assign an issue, and get back a merge request that has been tested, scanned and reviewed. No CI job to write, no webhook listener to host, and nothing to add to each project.
A job in your pipeline, or an agent with the pipeline built in.
Both open merge requests your team reviews. The difference is how much you build and look after.
| Claude Code GitLab CI/CD | Nightsquire | |
|---|---|---|
| Setup | A job in each project's .gitlab-ci.yml and a masked CI/CD variable | An OAuth application and a bot, once for your organisation. Enable a project and its webhook and bot membership are set up for you |
| Starting from a comment | A webhook listener of your own that calls the pipeline trigger API | Built in: assign the issue to the bot, or mention it |
| Where it runs | Your GitLab runners, on your CI minutes | Executors that serve only your organisation, managed in the EU or on your own hosts |
| Credentials | The API key and a GitLab token are variables of the job the agent runs in | Each step gets a placeholder. The real token and key never enter the container |
| Checks before review | Whatever you script into the job | Your tests, your pipeline with fix rounds, Semgrep and Gitleaks, and a separate security review |
| Models | Claude API, Amazon Bedrock or Google Cloud | Nightsquire AI, your Claude API key, or any OpenAI-compatible API |
| Seeing what happened | The job log | A dashboard with live logs, the agent's transcript, the hosts each step reached, tokens and cost |
| Status | Beta, maintained by GitLab | Closed beta |
Claude Code GitLab CI/CD as documented by Anthropic in October 2026.
Everything around the agent, ready on day one.
The part that takes the longest to build yourself is the part Nightsquire is made of.
Triggers that just work
Assign an issue to the bot or mention it in any enabled project. No webhook listener to write or host.
Your key, out of the agent's reach
The agent's container only ever holds a placeholder. The gateway swaps in your Claude API key on the way out.
Checks before anyone reviews
Your tests, your pipeline with fix rounds, Semgrep and Gitleaks, and a separate security review, on every run.
Every project in one click
Enable a project and its webhook and bot membership are set up for you. Nothing to add to its CI configuration.
A dashboard for every run
Live logs, the agent's transcript, the hosts each step reached, and the tokens and cost of the run.
Claude, or any other model
Use your Claude API key, Nightsquire AI in the EU, or any OpenAI-compatible API, per organisation.
What Claude users ask.
Is Nightsquire made by Anthropic?
No. Nightsquire is made by an independent Irish business and is not affiliated with Anthropic or GitLab. It can use Claude through a Claude API key of your own.
Can we use our own Claude API key?
Yes. Add it once for your organisation. It is encrypted with your organisation's key, and requests go from the executor's gateway straight to Anthropic, under Anthropic's terms. The agent's container only ever holds a placeholder.
Does Nightsquire read CLAUDE.md?
Yes. It reads each project's CLAUDE.md, and if a project has none, mention the bot in an issue and it suggests one as a draft merge request.
Do we have to use Claude?
No. Nightsquire AI works without any setup and runs in the EU. Or use any OpenAI-compatible API, including a model you host yourself on vLLM or Ollama.
What does it cost?
Pricing is coming soon. With your own Claude API key, the model's usage is billed to you by Anthropic, and every run shows its tokens. Nightsquire is in a closed beta; leave your email and we'll tell you when it opens.
Bring your Claude key. Skip the YAML.
Nightsquire is in a closed beta. Leave your email and we'll tell you when it opens to more teams.