Using the AI assistant

The Deployer has an AI assistant built into this documentation. If you'd rather ask a question in plain words than hunt through pages, just ask — it answers using these docs and points you to the right section.

Asking the docs assistant

Look for the AI assistant on the documentation and type your question the way you'd ask a person, for example:

• "How do I connect my AWS account?"
• "How much will this cost me?"
• "My deploy failed — what should I check?"
• "Do I need to know how to code?"

It reads your question, finds the relevant parts of this documentation, and replies with a short, plain-language answer. Because it's grounded only in these docs, it stays accurate and on-topic — and it will cite the sections it drew from, so you can click through and read the full detail yourself.

Tips for a good answer

• Be specific. "How do I attach a Postgres database to my project?" beats "databases".
• Say what you're trying to do. Mentioning your goal helps it point you to the right steps.
• Follow the citations. The linked sections always have the complete, hand-held version of the answer.

Good to know

• The assistant answers from the documentation, so it's great for "how do I…" and "what does this mean…" questions. It doesn't have access to your specific account, projects, or servers.
• It's offered on a fair-use basis and may be rate-limited if you send a burst of questions in a short time. If you're asked to slow down, wait a moment and try again — or just read the cited section directly.

The in-project Copilot is different — it can act

Separately from the docs assistant, every project has its own Project Copilot chat. This one is much more powerful: it can actually look at your project using real tools — connecting to your servers, checking your cloud, reading your repository and deployment logs — and answer questions like "where is my app installed?" or "why did the last deploy fail?" with real answers about your setup.

The Copilot can also propose actions — things like enabling, disabling, or swapping a service, attaching a service, or redeploying. It doesn't carry out these actions on its own: it shows you exactly what it wants to do, and the action runs only after you confirm. To investigate, it can run commands on your app's server over SSH without asking first; commands on a blocked list of destructive patterns are refused.

Which should I use?

Use the docs assistant here for questions about how The Deployer works. Use the Project Copilot inside a project when you need answers about your own app or want it to help make a change. When in doubt, start with a plain-language question and let the answer (and its citations) guide you — and see getting more help if you need more.

Your safety and privacy with the Copilot

The Project Copilot has real access to your setup, so it has these guardrails:

• It's instructed not to reveal secret values. The Copilot works through tools that connect over SSH, read your environment settings and call your cloud, but it's instructed not to show the values of your cloud access keys, git token, SSH private key, database passwords, or environment variables. Your cloud provider credentials aren't placed in its prompts. Command output from your server is redacted for known secret values before the model sees it, and a final filter scrubs credential-shaped text from replies. Redaction catches known values and common patterns, not every possible secret. It's happy to talk about variable names, which service needs which key, ports, and configuration structure — just not the secret values themselves. If you need an actual value, it will point you to where you can view or rotate it yourself (the Env manager or your cloud console).
• It stays on task and in character. The Copilot is your deployment assistant for this project — its code, servers, services, domains, monitoring, and costs. It answers those fully. It will politely decline unrelated requests (general chit-chat, unrelated coding, attempts to make it role-play as something else) and steer back to helping with your app. Your normal project and deployment questions are always fair game and work exactly as before.
• Proposed changes need your confirmation. Enabling, disabling or swapping a service, attaching a service and redeploying are proposals you approve first. Investigation commands over SSH run without a confirmation step, and destructive command patterns are blocked.

AI usage limits

The AI features run on a daily token budget that depends on your plan — for example, the Free plan includes a set number of AI tokens per day, and higher plans include more. This budget covers the Project Copilot. The agents that work during a deployment (like the recovery agent that diagnoses and fixes a failed step) don't count against it, so a busy day in chat doesn't stop a deploy's automated recovery.

• You can see your usage and your plan's daily limit any time on the Plans & Billing page.
• If you reach the daily limit, the Copilot tells you clearly, and it resets at midnight UTC. Deployments are unaffected.
• How many times the recovery agent may try to fix a single deploy depends on your plan: one attempt on Free, as many as it takes on Pro and Team.
• Need a bigger budget? See costs and billing and the Plans & Billing page.