Chat and Copilot problems

Use this page when the chat panel shows an error, a turn does not finish, or Copilot seems to lose context.

Chat shows a Copilot error

You see an error bubble in chat, Copilot needs your attention, or a message such as "GitHub Copilot is not signed in. Sign in to Copilot in Fabricator, then retry your message.", "Session not found", "Execution failed: Error: Session was not created with authentication info or custom provider", "rate limit exceeded", or a model access error.

Fabricator sends your prompt to the bundled GitHub Copilot engine. Authentication, model access, policy, rate-limit, and session errors come from that engine.

1. If the error asks you to sign in, select Sign in to Copilot in chat and complete the browser or device-code flow.
2. Retry the message yourself. Fabricator keeps your displayed conversation and draft, but does not automatically replay work after a failed or partially started turn.
3. If the model was marked Unavailable in the Model menu, open the model picker and choose Auto or another listed model.
4. If models do not load, sign out and back in to Copilot from setup or the chat notice.
5. If the error mentions rate limits or unavailable service, wait a few minutes, choose Auto or a fast model, and retry.
6. If repeated "Session not found" errors continue on the latest version, start New chat. This clears the conversation view, not your app files.

If the same error repeats, Report a problem.

A chat turn seems stuck

The assistant appears to keep working, the work log stops changing, or you cannot tell whether a turn is still active.

Long turns can run for many minutes. Fabricator streams the work log, command output, file edits, and the final "Worked for …" summary when the engine reports the turn is idle.

1. Expand the work log and check the latest step or command output.
2. If you know what should happen next, type a short steering message while the turn is running. Fabricator queues it for the active session.
3. If the turn is clearly wrong or waiting forever, select Stop.
4. After stopping, send a smaller follow-up prompt that names the file or problem to continue from the current project state.
5. If the preview or deploy is also running, wait for it to finish before starting another large request.
6. If Fabricator stops responding, restart it. Your project files and saved chat history are stored on disk.

If stuck turns happen repeatedly, turn on Settings → Diagnostics → Full diagnostics, reproduce once, then Report a problem.

The conversation disappeared or a new session started

You see an explanation that Fabricator started a new Copilot session, or chat history looks different after sign-in, restart, or New chat.

Fabricator keeps displayed chat history separately from the Copilot engine session. When the saved engine session is unavailable, Fabricator creates a new one so you can continue. New chat intentionally clears the visible conversation after confirmation and does not change your app files.

1. Update to the latest Fabricator version if this followed a "Session not found" error.
2. Continue in the visible chat if the history is still shown. Copilot can read the project files even if the engine session changed.
3. If history was cleared by New chat, describe the current task again or refer to the files that already changed.
4. Use the Code and History views to inspect or restore app file changes.

If history disappears without using New chat, Report a problem.

Copilot credits run out quickly

Your GitHub Copilot plan shows included credits or premium requests being used faster than expected during Fabricator work.

Fabricator uses your own GitHub Copilot entitlement. Each chat turn, deep review, design option, model selection, image attachment, and retry can use Copilot capacity. Fabricator cannot increase your plan limits.

1. Use the Model menu and choose Auto or a model marked Fast when you do not need a deep implementation pass.
2. Leave Reasoning effort on Auto or a lower effort for small edits.
3. Break broad requests into smaller prompts with clear acceptance criteria.
4. Avoid repeated Try again or Retry on the same large prompt; steer with a shorter correction instead.
5. Use quick checks and local inspection before running deep Copilot reviews.
6. Attach screenshots only when visual context is necessary.

If usage looks unrelated to your prompts, collect diagnostics and Report a problem. Diagnostics do not include your Copilot billing data.

Rayfin CLI reports missing dependencies

Fabric sign-in, a deploy, the preview, or a workspace list fails with a Node error such as "Cannot find package" or "Cannot find module" that names a package like rxjs, commander, or jsonwebtoken.

The project's node_modules folder is incomplete. A package install that stops part way, for example when creating a project times out, leaves package folders empty, and a sync or antivirus tool can remove files from the folder.

Fabricator checks for missing or incomplete packages when you open a project and before it deploys or signs in to Fabric, and reinstalls them. When a Rayfin CLI command fails because a package is missing a file, the log says "Some of this project's packages didn't install completely", and Fabricator reinstalls that package at its next check.

1. Select Redeploy, or Sign in to Fabric again. Fabricator reinstalls the incomplete packages first.
2. If the error comes back, close Fabricator and delete the project's node_modules folder. Then open Fabricator and the project again, and Fabricator installs all of the project's packages.

If the error keeps coming back, Report a problem and include the exact error.