Your agents interact with Fly.io in different ways: APIs, CLIs, reading docs, and even through the UI. Sometimes, they get stuck, notice gaps or bugs, and may have improvements to suggest. Until now, in the best case, they would report any of those to their humans. Most of the time, though, the issue probably got ignored, or the agent just found a workaround on its own.

Agents can now report feedback directly through the Fly API (and soon through the CLI, with the fly feedback command). Each report must point to either a specific request (using the fly-request-id header from the response) or a URL (for docs, the dashboard, etc.). The possible categories are:

bug: something went wrong or behaved unexpectedly
documentation: issues with the docs
feature_request: requesting a missing feature
other: everything else

Here is how a request to the API looks like:

curl -X POST https://api.fly.io/api/v1/feedback \
    -H "Authorization: Bearer $FLY_API_TOKEN" \
    -H "Content-Type: application/json" \
    -d '{"category":"bug","note":"...","request_id":""}'

You can find all the details on how to use the feedback endpoint in the API docs: https://docs.fly.io/reference/feedback-api . Additionally, our AGENTS.md file also contains instructions for agents too.

Since the feedback endpoint was released, we’ve already gotten some interesting feedback and the team fixed a few things based on them, like:

An agent using Kiro couldn’t connect to the Sprites MCP server because OAuth client registration kept rejecting its default auth method. Now it’s accepted.
An agent following our agent-ready guide hit a vague machine-limit error on a fresh org. The real issue was a missing credit card, and now the error actually says so.

The volume of reports is manageable so far, so the process of reviewing and triaging is still manual. As the number of requests grows, we’ll probably need to automate it somehow.