Failed answers

When the assistant cannot answer, the conversation shows a block that says in one sentence what happened, with Ask again and Open the assistant settings. The server's own words are behind What the server said. Each answer also has a link to its log: what was sent and what came back. This page lists every message and what to do.

The messages

The block saysWhat to do
"No assistant is chosen yet. Pick one in the toolbar or in Settings ▸ Assistant. Nothing was sent."choose who answers: Choosing who answers
"The cloud assistant has no key. Add the key in the assistant settings."paste the key: Keys and where they are kept
"The server did not accept the key. Check the key in the assistant settings."the key is wrong, revoked or of another provider; paste a new one
"Your server asks for a key. Paste it in the assistant settings under “Key for this server”."your own server wants a key: paste it under Key for this server (optional)
"The server does not have this model, or it was retired. Choose another model in the assistant settings."pick a model from the server's list
"The server asks to wait. Ask again in a minute." / "The server asks to wait 30 s. Then ask again."a rate limit or an empty account; wait, or check your account with the provider
"Something broke on the model server. Try again a little later."an error on the provider's or your server's side
"The model server does not answer. Check that it is running and the address in the assistant settings."start the server; check Server address
"The server answered, but not as a model. Check the address in the assistant settings."the address points to something else; for a local server it usually ends with /v1
"The server sent a reply that could not be read. Ask again; if that does not help, open the log of this answer."ask again; then look at the log
"A step could not be saved: the disk is full or the app cannot write its data file. The build stopped; what was built so far stays."free disk space
"The assistant could not answer. Ask again; the log of this answer shows what happened."ask again; then look at the log
"The assistant tried to change the graph, but the chat is in Ask mode. Switch to Build and ask again; nothing was changed."switch to Build
"The assistant asked for a tool that is not available here. Nothing was changed. Try asking in other words."rephrase; a small model may need a bigger one

Above an answer

  • "The model read only part of the request" and "The model's context window is almost full": see The context window.
  • Answered without tools: "This model cannot call tools, so the assistant answered from the graph it was sent: it could not look anything up, read more or change the graph. To build or to get drafts, choose a model that can call tools (for example qwen3 or llama3.1)."
  • If even that does not work: "This model cannot call tools, and without them the assistant cannot read the graph, look up your nodes or suggest changes, so it did not answer. Choose a model marked “tools”…"
  • "Stopped by you: the answer may be unfinished." You stopped it; its build card says Stopped.

Other things the chat says

  • If the window was closed during a build, the build is stopped and the chat says so at the next start: "The assistant's work stopped when the app was closed. What it had added stayed on the canvas (nodes: 4); Undo takes back consecutive assistant steps, stopping at your edits."
  • "This build is too old to undo in one step: the version of the graph before it is no longer kept." when you try to undo an old build.

A step that did not work

Inside a build, a single step can fail without stopping the build: "Could not add the node …" with "This node is not in your ComfyUI. Similar ones: …", or "The graph changed first, so this step was not applied." The assistant can try another way. After 3 failures in a row it pauses: see Pausing and stopping.

An odd answer

An answer that is wrong, invented or off the point is not shown as an error. Ask again in other words; ask about one object by right-clicking it; check that the model can call tools and has a large enough context window. To have it looked into, save the log: The conversation log.

See also

This page is imported from the application repository's wiki (docs/wiki/assistant/Failed-Answers.md, revision b8e3f71). The same guide is built into the app: Help ▸ User guide.

To refresh it, run npm run import and then npm run build in the website folder.