Help and troubleshooting
SkillPiece support
Set up your model, solve common capture problems, and collect the right details when something is not working.
Last updated September 5, 2026
Get started
- Open SkillPiece and choose Chrome's built-in model or add a custom provider.
- Run Test local model or Test connection. Read every result row before saving a provider.
- On an ordinary web page, select a useful passage. Choose Focus on selection for a temporary clean reader or Capture selection to keep learning from it.
- Follow Open Topic to watch complete Concepts and knowledge pieces appear as each source unit finishes. You can pause, leave, and resume safely.
- Open a Concept and choose Start rehearsing to review questions with answers shown, or Start explaining to practice in your own words.
Chrome's built-in model
The built-in model runs on your device and needs no API key. Chrome controls whether the model is available and when it downloads.
The model is downloadable or downloading
Keep SkillPiece open while Chrome downloads the model. Use an unmetered connection and make sure the volume holding your Chrome profile has enough free space.
If progress stops, return to Settings and use Try the download again. You can inspect Chrome's model state at chrome://on-device-internals.
The model is unavailable or the API is not exposed
First update Chrome. Then open chrome://flags and enable both Prompt API for Gemini Nano and Optimization Guide On Device Model with Enabled BypassPerfRequirement.
Fully quit every Chrome window and relaunch. If needed, open chrome://components, find Optimization Guide On Device Model, and press Check for update.
Chrome currently requires a supported desktop system, substantial free storage, and suitable RAM, CPU, or GPU capacity. Flags can help with detection, but they cannot make unsupported hardware run the model reliably.
The model is ready, but the capability test fails
Expand the test result and read the first failed row. The test performs real generation, structured-output, schema, and grading checks. A downloaded model can still be too weak for reliable grading on a particular machine.
You can switch to a custom provider without changing or deleting your local SkillPiece library.
Custom providers
Choose the provider type first, then enter a recognizable name and the exact model identifier. Add credentials only when the selected endpoint requires them, and use the endpoint's exact base URL.
- Official hosted endpoints: choose the matching provider type so SkillPiece sends the correct request format.
- Local Ollama: no API key is required. The usual base URL is
http://localhost:11434/v1. Restart the server withOLLAMA_ORIGINS="chrome-extension://*" ollama serveif the test reports that Chrome extensions are blocked. - Other compatible endpoints: choose OpenAI-compatible and enter the endpoint's Chat Completions base URL. Add a key only when that endpoint requires one.
- Remote endpoints: HTTPS is required. Plain HTTP is accepted only for loopback addresses such as localhost.
During onboarding, Save and use provider saves and activates the profile. In Settings, Save provider stores or updates a profile without switching away from the currently active model; select the round check-mark control beside a saved profile to use it. API keys stay in local extension storage and are excluded from exports and diagnostic reports.
Focus and Capture problems
The selection actions do not appear
Confirm that Quick capture is enabled in Settings, then select a meaningful passage on an ordinary website. Chrome prevents extensions from drawing on protected pages such as chrome:// pages and parts of the Chrome Web Store.
If the page was already open when SkillPiece was installed or updated, reload that page once. The extension popup can still open the clean reader with Focus or capture the readable page with Capture this page.
Capture opens model setup
First-run setup has not been completed. Press the action, confirm a working model in the welcome screen, and finish with Continue.
The action says Fix model setup
Your active model cannot currently accept work. Press the action to open Settings, then repair or replace the active provider and run its capability test.
Failed generation or grading
Capture extraction and grading begin only from the corresponding action. If Chrome terminates the extension service worker while requested work is still running, SkillPiece may resume that work from the durable queue so it is not lost. A terminal failure is never retried automatically; use the visible retry action after confirming that the active model passes its test. Timers never start speculative model work or retry a terminal failure. A slow grade shows progress instead of presenting self-grading until the request actually fails.
SkillPiece ties waiting work to the provider setup selected when the action began without copying the key, endpoint, or model name into that work. If you switch or edit the selected provider before the request is sent, nothing is sent and the work is marked failed. Press Retry to make a new request with the model selected then.
If the same failure returns, open Settings → Advanced. The diagnostics section shows failed jobs and can export a scrubbed report containing counters and allowlisted error categories, never raw errors, captured text, answers, or provider keys.
Before asking for help
Please include:
- Your SkillPiece version from
chrome://extensions. - Your Chrome version and operating system.
- The exact status or error text shown by SkillPiece.
- The steps that produced the problem and what you expected instead.
- The diagnostic report from Settings, if you are comfortable sharing it.
Protect your library before resetting
If troubleshooting requires a reset, export your full library first from Settings. Removing the extension or erasing SkillPiece data permanently removes the local library from that Chrome profile.
Backup exports exclude provider keys and replace raw provider or job error text with fixed local text. On import, pending or running jobs and pending grades are restored as failed and never run automatically. Full replacement and Start over also invalidate and clear unsubmitted recall drafts and transient capture or reminder state from the prior library. Use the visible Retry or self-grade action when you are ready.
Read the privacy policy for complete details about local storage, remote providers, exports, and deletion controls.