When something does not work

This page lists the messages the app shows when something goes wrong. Each one says what you see, why it happens, and what to do. The messages are grouped by what you were doing.

Many messages appear on the Activity page. Open it from the Agents page. Press a job to read its summary: when a job stops, the summary says why.

A part of a message in "…" changes each time, for example a number or a file name.

Signing in

"Wrong e-mail or password."

• Why: the e-mail address or the password does not match an account.
• What to do: check both and try again. If you forgot the password, press Forgot your password?. If you made your account with GitHub, press Sign in with GitHub instead.

"Confirm your e-mail address first: open the link we sent you, or ask for a new one."

• Why: you made your account with an e-mail address, and the address is not confirmed yet.
• What to do: open the confirmation link in the e-mail that we sent you. To get a new link, press Send the confirmation link again.

The e-mail does not arrive

This can be the confirmation link, the link to choose a new password, or an invitation to a workspace.

• Why: the e-mail can be late, or your mail program can put it in the spam folder.
• What to do: wait a few minutes, then check your spam folder. Or sign in with GitHub instead: press Sign in with GitHub. If GitHub has the same e-mail address, you get the same account.

"Too many wrong passwords. Wait 15 minutes."

• Why: too many wrong passwords came from your network in a short time.
• What to do: wait 15 minutes, then try again. Or press Forgot your password?.

"That sign-in didn't work"

The page then says why. Here is what to do for each reason:

"That link didn't work"

• Why: a link from an e-mail works once, and only for a limited time. This link is old, or was used already.
• What to do: sign in, and ask for a new link if you need one.

"Sign-up is by invitation for now"

You can also see "Sign-up is by invitation. Ask a workspace admin to invite you."

• Why: new accounts can be made only with an invitation.
• What to do: open the invitation link that you got, or ask a workspace admin to invite you.

Connecting GitHub

"Only a workspace admin or owner can connect GitHub."

• Why: your role in the workspace cannot connect GitHub.
• What to do: ask a workspace admin or owner to connect it.

"The app reaches no repositories yet. Choose some on GitHub."

• Why: the InfraStudio AI app is installed on GitHub, but it may use no repository.
• What to do: press Change on GitHub and choose the repositories that the app may use.

"Every repository the app reaches has a product. To add one, change the app's repositories on GitHub."

• Why: each repository that the app may use is in a product already.
• What to do: press Change on GitHub and choose more repositories. Also check that you installed the app on the right account or organization.

"suspended on GitHub"

• Why: the app is suspended on that GitHub account or organization. The agents cannot reach its repositories.
• What to do: on GitHub, open the account's or the organization's settings for installed apps, and remove the suspension.

"A product in this workspace uses that repository already."

• Why: one repository can be in only one product of a workspace.
• What to do: open the product that has it, under Settings → Products.

"The … plan has room for … active products. Archive one, or upgrade on the Billing page."

• Why: your plan allows only this number of products.
• What to do: put a product away in its settings, or choose a larger plan on Settings → Billing. See Plans and limits.

Saving your AI key

"Only a workspace admin or owner can change keys."

• Why: your role in the workspace cannot add or change keys.
• What to do: ask a workspace admin or owner to add the key.

"The agents are offline, so keys can't be changed or tested now."

You can also see "The platform cannot store keys safely right now. Nothing was saved; try again in a moment."

• Why: the agents are not reachable at this moment, so a key cannot be saved.
• What to do: wait a few minutes, then try again. If this lasts longer than an hour, write to us.

"That isn't an Anthropic API key. An API key comes from the Anthropic Console and starts with sk-ant-api."

• Why: the key that you pasted is not an Anthropic API key. A key from a Claude subscription, for example, is not an API key.
• What to do: make an API key in the Anthropic Console and paste that one. See Add your AI key.

"There's no AI provider key for this project. Add one on the Integrations page: …"

You see this as the reason why a job stopped, on Activity.

• Why: the product has no key of its own, and the workspace has no key either.
• What to do: add a key on Settings → Integrations. A key of the workspace is shared by every product that has no key of its own. See Add your AI key.

"The project runs on …, but there's no key for it. Add it on the Integrations page."

• Why: in the product's settings, the AI provider is set to one provider, but there is no key for that provider.
• What to do: add a key for that provider on Settings → Integrations. Or, in Product settings, set the AI provider back to "Automatic", so the first provider with a key is used.

"The agent's … failed: …"

The end of the message is your AI provider's own answer. It can say that the key is wrong or expired, that your account has no access to the model, or that you used up your quota or reached a rate limit.

• Why: your AI provider refused the request.
• What to do: check your key, your model and your usage on your AI provider's website. In Product settings, check that the model names are ones that your account can use. For Azure OpenAI, the model is your deployment name. A larger InfraStudio plan does not change your provider's quota.

Starting an agent

There is no Start now button

• Why: there are three reasons:

  • Only a workspace admin or owner can start agents by hand.
  • Mentions and Auto-merge cannot be started by hand. They only answer what happens on GitHub.
  • The product is set so that the platform does not run its agents.
• What to do: ask a workspace admin or owner. For the last reason, turn the platform on in Product settings.

"The agents are offline"

• Why: the agents are not reachable at this moment. Nothing would happen if you started one.
• What to do: wait a few minutes, then reload the page. If this lasts longer than an hour, write to us.

"Which pull request? Its number, such as 42."

• Why: this agent works on one pull request, and the number is missing or is not a number.
• What to do: type the pull request's number into Pull request number, without other text. The number is in the pull request's address on GitHub.

"A job of … is still on its way. Wait for its answer."

• Why: you started this agent a moment ago, and that job has not started yet.
• What to do: wait for its answer on the Agents page, or on Activity.

A job stays "In line"

• Why: your plan allows only a few jobs at the same time. The job waits until another one is done.
• What to do: wait. It starts by itself. See Plans and limits.

"This month's jobs are used up, so the agents are paused until the 1st."

• Why: your workspace used all the jobs of its plan for this month.
• What to do: wait until the 1st of the next month, or choose a larger plan on Settings → Billing. Nothing is charged automatically.

"… timed out after … min"

• Why: each plan limits how long one job may take. This job took longer, so it stopped.
• What to do: ask for a smaller piece of work, or choose a larger plan. See Plans and limits.

"This run reached its plan's limit of … AI calls."

• Why: each plan limits how many AI calls one job may make. This job reached the limit, so the AI stopped answering it.
• What to do: ask for a smaller piece of work, or choose a larger plan. See Plans and limits.

"(refused: …)" at the end of a summary

• Why: the agent tried to use a tool or a command that it may not use, and the platform stopped it. The part after "refused:" names what was stopped.
• What to do: usually nothing. The agent goes on without it. If the job could not finish, ask for the work in smaller steps. If your own instructions in the repository ask the agent to run commands, see Write your own instructions.

"The product is archived, so no agent works on it. Unarchive it in its settings first."

• Why: the product was put away, so its agents do not work.
• What to do: bring the product back in its settings, under Settings → Products.

Nothing happens after a mention on GitHub

• Why: the agents answer @infrastudio-ai only for people with write access to the repository. A comment from anyone else is ignored. Mentions must also be on.
• What to do: ask someone with write access to write the comment. Check on the Agents page that Mentions is on.

"CI hasn't failed on the latest commit (…), so there's nothing to fix yet."

The agent writes this on the pull request after @infrastudio-ai fix.

• Why: the automatic tests did not fail on the newest version of the pull request, so there is nothing to fix.
• What to do: wait until the tests have run. Use @infrastudio-ai fix when they fail.

A suggestion

"the change was not pushed: no changes"

• Why: the agent finished, but it changed no file. Often the task was not clear enough, or the change was made already.
• What to do: read the job's summary on Activity. Then start it again with a note that says more exactly what to change. Or say no to the suggestion.

"the change was not pushed: …: CI workflows"

The end can also name other kinds of files, for example "secrets and environment files", "dependencies" or "database migrations". It can also say "blocked by the project's settings (…)".

• Why: the change touches a file that the agents may never change. Some of these rules are the same for every product. The others are your own, in Extra paths agents may never touch in Product settings.
• What to do: make this part of the change yourself. You can ask the agents for the rest. See Repository rules and privacy.

"too big, so nothing was pushed: … files changed (limit …)"

The message can also name lines: "… lines changed (limit …)".

• Why: the change is larger than the product's limit for one change. The limits are Most files per change and Most lines per change, in Product settings under Repository rules.

• What to do: on the suggestion, choose one of these:

  • "Use the large change anyway" opens the pull request as it is, up to twice your limit. Nothing runs again, so you pay nothing more.
  • "Split it into smaller changes" makes a new plan in smaller parts.
  • "Start again" makes the change again from the beginning. It is a new job.

"The automatic tests failed"

• Why: the tests in your repository failed on the pull request.
• What to do: open the suggestion and read why. You can ask the agents for a change, or fix it yourself on GitHub.

"Pull request #… clashes with changes that are now in the main branch of your code."

• Why: someone changed the same code after the agents made their change.
• What to do: start the change again, so that it is made on today's code.

"A second AI read the change and asks for changes"

• Why: after the tests pass, a second AI reads each change. It found a problem.
• What to do: read what it said on the suggestion. Ask for a change, or put the change in anyway with "Put it on … without the second reading".

"… cannot take “…” now, because its state is: …"

• Why: the suggestion is in a stage where this action does not fit. For example, you cannot merge a suggestion that has no pull request yet.
• What to do: wait until the suggestion reaches the right stage, or choose another action.

The result is not what you expected

• What to do: read the job's summary on Activity and the review on GitHub. Then ask for a change, or say no to the suggestion. Start with a smaller and more exact task.

Auto-merge

"Auto-merge can go live only once every ADLC check passes."

• Why: Auto-merge can put changes into your code by itself only when your repository passes its checks. For example: the default branch is protected, the automatic tests pass, the deploy runs, the health address answers, and Auto-merge has a track record in Shadow mode.
• What to do: in Product settings, switch Auto-merge on in Shadow mode. Let the platform check the repository. Then fix what the Auto-merge checks list there. See Turn on Auto-merge.

Auto-merge turned itself off

• Why: after a change went on your product, your product failed its health check twice. Auto-merge then turns itself off, and opens a pull request that takes the change back.
• What to do: on the Agents page, read why it turned itself off. Merge the pull request that takes the change back, or fix the problem. Then turn Auto-merge on again.

Auto-merge does not merge a large change

• Why: a change that you accepted as larger than your limit always waits for a person.
• What to do: review it and merge it yourself.

Notices about the agents

These notices appear in the side bar of the app.

"No news from the agents for …"

• Why: the agents have not reported for a while. Work can be late.
• What to do: open Activity to see what is in progress. Wait a few minutes. If the notice stays longer than an hour, write to us.

"No answer from Telegram for … min"

• Why: the platform did not get an answer from Telegram for a while. Messages in Telegram can be late.
• What to do: wait a few minutes. If the notice stays longer than an hour, write to us.

"The CI checks last ran … min ago"

• Why: the platform has not checked the automatic tests of your pull requests for a while. News about tests and merges can be late.
• What to do: wait a few minutes. If the notice stays longer than an hour, write to us.

"The agent reports a problem."

• Why: the agents found a problem on the platform's side.
• What to do: wait a few minutes. If the notice stays, write to us.

Write to us

If your problem is not on this page, or a notice does not go away, write to us. Our e-mail address is at the bottom of this page, and on the About page. Say which product it is, what you did, and the message you see.