AI prompts in JupyterLab notebooks
Manual · Previous: Saved notebooks and privacy
| Symptom | What to do |
|---|---|
| Configure AI or + AI Prompt is missing | Open a notebook, confirm installation in the server environment, restart the whole server, and refresh the page. |
| Server endpoint unavailable / 404 | Try Retry in Configure AI. If you just installed or updated, restart the whole server; browser reload alone may leave the server component unloaded. |
| Provider unavailable / Run disabled | Save that provider’s key, or select one already configured. |
| Name is not defined | Run the Python cell defining the referenced variable or function in this notebook’s kernel. |
| AI misses your notes | Select the intended question, check its Context mode and cell choices, use Details → Check context if you want a fresh estimate, and inspect partial/omitted feedback. Rerun with Keep answer off after editing. |
| AI misses a plot or code output | These are not currently included; add a text explanation to a Markdown cell above the prompt or to the prompt itself. |
| AI asks questions when you want a direct answer | Choose Compact or Full in the notebook defaults or the cell’s Override controls, then run it again. |
| One cell ignores changed notebook defaults | Check its Override controls. Choose Notebook default in the provider menu to clear its provider, model, and effort choices, or Use notebook defaults to clear its provider, model, style, and effort choices. |
| A completed AI prompt will not run again | Turn off Keep answer on that prompt. Protected prompts are skipped by Shift+Enter. |
| Later answers still contain a mistake you corrected above | Rerun the affected prompts with Keep answer off. Editing earlier text does not update existing later answers automatically. |
| A prompt ignores the notebook’s Keep AI answers toggle | Reset that cell’s explicit Keep answer choice to inherit the notebook default. |
| Run All does not run AI prompts | Confirm version 0.1.5 or newer, restart the server after upgrading, and check Keep answer. Headless notebook execution does not activate the extension’s frontend hook. |
| A variable is missing after restarting and running all | A kept AI answer did not repeat its old function calls. Recreate the variable in a code cell or deliberately rerun the AI prompt that created it. |
| A tutor conversation keeps starting over | Put your reply in a new AI Prompt below the tutor’s answer. Rerunning the original prompt replaces that exchange. |
| Code will not copy | If the browser blocks clipboard access, select the code text and copy it manually. |
| Old answer disappeared after rerunning | Reruns replace the paired answer. Copy text into an ordinary Markdown cell beforehand to preserve another version. |
| Model unavailable / key rejected / quota reached | Check the selected connection and model. For ChatGPT, reconnect if needed or wait for the displayed reset time; your selection is preserved. For API mode, check the key and that account’s access or quota. |
Current size limits are deliberately bounded:
| Item | Limit and behavior |
|---|---|
| Request context | A 64,000-character host estimate for each submitted question or notebook-tool round, including tool definitions and message framing. Tools, instructions, the expanded question, and completed tool results take priority; remaining space goes to nearest selected source and complete earlier AI pairs. The ChatGPT runtime may make internal inference or recovery requests within a round; this limit does not account for those internal calls. |
| Notebook snapshot | Up to 10,000 ordered cells as a transport safety limit. All earlier eligible Markdown/AI questions are scanned for declarations before text selection. |
| Source at the context boundary | Source may contribute only its ending above or beginning below, labeled partial. AI history pairs are never split. More distant material is omitted. |
| Current prompt | Up to 16,000 characters before live-value substitution. |
| Live references | Up to 20 distinct names combined: current-question variables plus all inherited/current tools. |
| Variable representation | Up to 2,000 characters per value. |
| Function result | Up to 4,000 characters per tool result. |
| Notebook-tool rounds | Default 5; configurable from 0 to 10 in nbinlineai’s JupyterLab settings. One returned group may contain several declared tool calls; internal ChatGPT inference does not count as another notebook-tool round. |
This version supports text prompts and Python kernels. It does not automatically include rich outputs or images or execute generated code.