Choose how summaries and Q&A are generated. Local runs Gemma in your browser via WebGPU — first load downloads ~1.5 GB of weights, cached after.
Adds the option to enrich Ask answers with recent web results. Without this, Ask uses only the report itself. All providers support free tiers; key stays in this browser.
The header pill controls saving the index + the texts you've opened. Optionally, also save your AI-generated summaries and chats alongside, so they're part of your portable folder. Off by default — your AI work stays in this browser's IndexedDB unless you enable this.
Files land under <folder>/<corpus>/ai/. Markdown for summaries, JSON for chats. Delete from Finder/Files to remove — nothing else clears them.
📘 New here? Open the full user guide → — covers AI setup, BYOK costs/limits, web search enrichment, privacy details.
A single-file browser app for India's parliamentary and executive record — eight searchable corpora across four groups:
Each corpus is scraped on its own GitHub Actions cron and served as static JSON from Cloudflare Workers — no server, no auth, no API key.
Open any report and you get four tabs: Details (metadata + PDF links), Full text (when extracted), AI summary (one click → plain-English briefing), and Ask (chat with the report). AI runs on your device by default — open Settings (the gear icon, top right) to load a model or set a BYOK key. Full walkthrough in the user guide.
The corpus chip strip near the top groups the eight datasets under four labels (oversight, legislation, proceedings, executive). Click any chip to switch the list, header stats, and filter row to that corpus. Press ⌘K for the cross-corpus search palette.
The filter row is corpus-aware — DRSC offers committee + Lok Sabha + category; CAG offers ministry + sector + year; Gazettes offers category + ministry + language; and so on. Search always works as substring match against titles by default; enable full-text search for a corpus in Settings to pre-fetch its body content and search inside.
Runs an open-weight model entirely in your browser via WebGPU. Five options today: Gemma 4 E2B (~1.5 GB, default), Gemma 4 E4B (~4.9 GB, stronger), Ternary Bonsai 1.7B (~470 MB, smallest), Bonsai 4B (~1.1 GB), and Bonsai 8B (~2.2 GB, 64K context). First load downloads weights from Hugging Face; subsequent loads are instant from browser cache. Requires a recent Chrome/Edge/Brave or Firefox 130+ with a compatible GPU. No data leaves your device.
If WebGPU isn't available or you want a stronger model, configure a provider in Settings. Free options:
ollama pull llama3.2 then enable CORS for browsers.Paid: Anthropic (best summaries) at console.anthropic.com, OpenAI at platform.openai.com.
AI summary + Ask are available in DRSC, CAG, FC, Bills, LC, Debates, Gazettes — whenever the report has extracted text (rows with the text badge). For metadata-only rows the AI tabs show a hint to open the source PDF instead.
Questions is intentionally metadata-only here — its records are short Q&A pairs without long text bodies. Per-MP and per-question analytics will live in Netas Explorer (sibling project, coming soon).
localStorage, never sent anywhere except your chosen provider)IndexedDB)sansadsaar-data.naklitechie.com (DRSC, CAG, FC, LC, Bills)sansadsaar-proceedings.naklitechie.com (Debates + Questions)sansadsaar-gazettes.naklitechie.com (Central Gazette)huggingface.co — model weights, only on first Gemma load.index.html from Cloudflare Workers Static Assets, on the custom domain.Chirag Patnaik — SansadSaar is part of a series of single-file browser-native tools.
All projects · chiragpatnaik.com · naklitechie.com · Substack
Scraping logic, committee map, and the original idea — all from ParliamentWatch by Pranay Kotasthane.
The idea of treating India's Central Gazette as a queryable corpus comes from egazette by Sushant Sinha. SansadSaar's gazettes module is independently implemented (pulls from archive.org's gazetteofindia collection, no code copied), but the conviction that this was worth doing at all is owed to his project.