FormPilot 1.2.0

Added

• Richer anonymous usage (same off switch, still nothing in Local-only mode): suggestions requested / shown / accepted (acceptance rate), how a suggestion was inserted (click, Enter, Tab, Alt+number) and its position, how the popup was opened (on focus, icon, Alt+J, Alt+↓, right-click menu, chat question), Regenerate / Shorter / Longer / Refine and the other popup buttons, keyboard shortcuts and context-menu items, the toolbar popup and settings pages, memory and provider actions, and learn-from-corrections prompts.
• Quality signals: the outcome of every suggestion request (AI, cache, saved answer, fact, empty, error, timeout), AI errors by category (rate limit, auth, timeout, network, locked…), latency buckets, response-cache hits, provider fallbacks, fact extraction and embedding results, provider tests.
• A daily snapshot of on/off settings and preset choices (never sites, prompts, persona text or models), the number of configured providers by type, and memory size buckets.
• Crash capture in the content script, toolbar popup, settings page and offscreen page, forwarded to the background with only the error class and FormPilot's own stack frames.

Changed

• popup_opened now counts each popup open once (it used to count both the instant lookup and the AI request).

FormPilot 1.1.0

Added

• Share anonymous usage (Privacy & Data), on by default. It shares which features are used, daily counts of filled fields by source (memory, history, AI), setup milestones, the AI provider type (Gemini, Groq, Ollama…) and error types, with a random id made on install.
• It never shares field contents, page addresses, memory, provider URLs, headers or keys, and it sends nothing in Local-only mode. Turning it off deletes what was sent.

Changed

• Asks for the alarms permission, used to send usage a few times a day.

FormPilot 1.0.0

First public release.

Added

• Suggestions under every field: a dropdown anchored to the focused field, with several options tagged Memory / History / AI / Cached, streaming output, and Regenerate · Shorter · Longer · Refine…
• The AI decides the answer length from the question and the field; it always respects maxlength.
• Whole-form fill in one request (Alt+Shift+J), with Undo. Passwords, cards, OTPs, CAPTCHAs and consent checkboxes are never touched.
• Memory: upload PDF, DOCX, TXT, MD, CSV, JSON or HTML (parsed locally), or write notes. Key facts are extracted automatically. Hybrid BM25 + embedding retrieval with reciprocal-rank fusion.
• Instant answers from key facts, saved answers and a response cache, with no AI call.
• Learns from corrections: type the right answer and FormPilot offers to remember it (Ask me / Save automatically / Off).
• Providers with automatic fallback: Google Gemini, Chrome built-in AI (Gemini Nano), OpenRouter free models, Groq, Ollama, LM Studio and any OpenAI-compatible endpoint. Includes a live model picker with context, pricing and size.
• Local Models page: one-click install and removal of open models through Ollama, with a catalog of specs and RAM needs.
• Popups and chat questionnaires: renders in the browser's top layer and mounts inside modals and drawers. Chat-questionnaire sites (Naukri by default) read the bot's latest question and refresh after each reply.
• Movable popup: drag it anywhere, with the position remembered per site or everywhere; reset it from the ⋯ menu or settings. Pin it to keep it on screen. Clicking a field reopens a closed popup. Placement, size, opacity and visible parts are configurable. Alt+1…9 inserts a suggestion directly.
• Password-protected API keys: keys are encrypted with AES-256-GCM using a key derived from your password (PBKDF2, 600,000 iterations). A 4-digit PIN is allowed, and wrong attempts are throttled. An optional strict mode locks keys after a browser restart.
• Privacy: sensitive fields are never read or filled, secrets are redacted before sending, and there is a local-only mode and per-site rules.
• Settings app: 12 pages, light and dark themes, logs and diagnostics, export and import, and an About page.
• Tests: 136 unit tests (Vitest) and 23 end-to-end tests (Playwright driving real Chromium with the extension loaded).