AI Assistance

Devpilot's AI assistant is a conversational helper attached to every deployment. Instead of reading through long logs or jumping between tabs, you can ask a question in plain language — "Why did this deploy fail?", "What does this error mean?", "What should I try next?" — and get an answer that uses the actual context of the deployment you are looking at.

What the AI can do

When you ask a question, Devpilot passes the AI assistant a rich snapshot of the deployment, including:

• The deployment itself — Status, progress, provider, repository, branch, commit ID, commit message, start and end times, duration, and the current process message.
• The app and project — Name and platform, so answers are specific to your stack.
• Every hook in the deployment — Name, stage, status, output, message, duration, and timestamps for each hook that ran (or was skipped).
• Collected errors — Messages and outputs from any hooks that failed.
• Console output — The deployment's log output.
• Recent successful deployments — Up to three prior successful runs for the same app, so the AI can compare what changed.

Because the assistant sees all of this context, it can:

• Explain what an error in a specific hook most likely means.
• Suggest a next step or fix based on the failure pattern.
• Summarize what happened during a deployment in plain language.
• Compare a failed deployment against the last successful one and highlight the differences.
• Answer follow-up questions — you can pass extra context along with each query.

Using the assistant

Choosing an AI model

Devpilot supports more than one AI backend. You can select which model to use when sending a query — the available options are managed at the workspace level. If you do not select a model, Devpilot falls back to a sensible default. The response includes the model that was used, so you always know which one generated the answer.

Limits and guardrails

• Query length. Each query is capped at 2,000 characters. Keep questions focused — the deployment context is already included automatically, so you do not need to paste logs or error messages into your prompt.
• Rate limiting. You can send up to 20 AI requests per minute per deployment. This protects the AI service and keeps response times fast. If you hit the limit, Devpilot returns a clear error asking you to wait.
• Safety checks. Every response is safety-checked and logged for auditability. If the AI service is temporarily unavailable, Devpilot returns a friendly error rather than hanging.
• Scope. The assistant answers questions about the deployment you are viewing. For workspace-wide questions — such as "Which apps fail most often?" — use Deployment Insights instead.

Tips for better answers

• Ask specific questions. "Why did the build hook fail?" is more useful than "What's wrong?".
• Reference hook names. If you mention a hook by name ("What happened in Run Migrations?"), the AI will focus on that hook's output and status.
• Use it alongside diagnosis. For failed deployments, run deployment diagnosis first for a structured root cause, then ask the AI follow-up questions for clarification or next steps.
• Iterate. If the first answer is too high-level, ask the assistant to go deeper on a specific step or error message.

Next steps