FilingForge
Health Uyari
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 5 GitHub stars
Code Basarisiz
- rm -rf — Recursive force deletion command in .github/workflows/release.yml
Permissions Gecti
- Permissions — No dangerous permissions requested
Bu listing icin henuz AI raporu yok.
Turn any Indian listed company's official BSE filings into clean, AI-ready Markdown — local-first, free, MIT, no telemetry. Hand your AI primary sources it can finally read.
FilingForge
Official filings, as clean Markdown your AI can actually read.
Search any BSE-listed Indian company → pull its official filings → get a clean, AI-ready Markdown library on your own machine.
Built by AlphaForge Integrated Solutions · works with Claude Desktop and Claude Code through its MCP server.
The problem: PDFs are where AI research quietly goes to die
Drop a filing PDF into the Claude desktop app or Codex and the tables collapse into mush, scanned pages become nothing, and you burn thousands of tokens on layout garbage. The model hands you a confident, wrong answer — and you never see that the context was broken.
FilingForge fixes the input. Every text-based filing becomes clean, structured Markdown, with an INDEX.md your AI reads first. Same model, finally fed properly.
A filing PDF → your AI The same filing, FilingForge
───────────────────────── ────────────────────────────
%PDF-1.4 … /F2 9 0 R ## Revenue (₹ cr)
Revenue 1,2 4 8.3 2 ,1 09.7 | Segment | FY24 | FY23 |
(cid:32)(cid:71) [table spans pages] |-----------|------:|------:|
running headers, footers, page nums | Retail | 1,248 | 1,109 |
…34,000 tokens of layout noise | Wholesale | 860 | 740 |
*Source: Annual Report FY24, p.42*
The download is the easy part. The transformation is the product.
📸 Screenshots
Look up any BSE-listed company — it resolves against the live exchange:

Your library — every company, every filing, indexed and AI-ready:

Skills — turn the clean library into finished work in your own AI:

⬇️ Install
Download the latest build for your platform from GitHub Releases.
Free. No account. Your data never leaves your computer.
macOS
FilingForge is Apple-notarized — download the .dmg, drag to Applications, and open it. No quarantine warnings, no workarounds.
Windows
Windows SmartScreen may warn about an unrecognized publisher. Click More info → Run anyway. The installer is a standard per-user NSIS package — no admin rights required.
⚙️ How it works
One window. Search a company, pick what you want, and it builds the library.
| 1 · Search | Type a company. FilingForge finds it on BSE and pulls its official filings — annual reports, results, investor presentations, and more. Annual reports go back to 1997 where BSE has them, including for companies since delisted. | |
| 2 · Convert | Every PDF becomes a clean .md sibling — readable, structured, AI-ready. No PDFs for your model to choke on. |
|
| 3 · Index | A per-company and master INDEX.md map every document, so you (or your AI) point at one folder and have everything. |
The library is laid out year-wise, and refreshes are smart and incremental — re-run a company and FilingForge pulls only what's new, leaving your existing Markdown untouched.
Two sources, one library. BSE's announcements feed only carries annual reports from 2015, when LODR Reg. 34(1) started requiring them. FilingForge also reads BSE's separate annual-report archive, which reaches back to 1997 — so a library covers two decades, not one, and a company that has since delisted keeps its history. Where a report sits in both places it is counted once, and refreshing an existing library adds the older years without re-downloading anything you already have.
FilingForge converts text-based filings (the vast majority of BSE documents) to clean Markdown — deliberately lightweight, no OCR, no GPU. The occasional scanned-image PDF is saved as-is and clearly flagged, so your AI never reads fabricated text.
🔌 MCP server — let your AI drive the library
FilingForge ships an MCP server, so an MCP-aware client (Claude Desktop, Claude Code, and others) can work the library directly instead of you copying files around.
pip install "filingforge-engine[mcp]"
filingforge-mcp --root ~/FilingForge
| Tool | What it does |
|---|---|
list_companies |
What's in the library, with filing counts |
get_index |
A company's INDEX.md, or the master index |
search_filings |
Find documents by title, narrowed by company / category / year |
read_filing |
Returns the clean Markdown — never the PDF |
pull_company |
Download a company from BSE into the library |
refresh_company |
Re-pull one already held, keeping its original categories |
Two deliberate constraints. The library root is fixed when the server starts, and every path is resolved and checked against it — a model cannot read outside your library. And pull_company never guesses between similarly-named companies: an ambiguous name returns the candidates and downloads nothing.
The MCP extra installs into its own environment —
mcp2.x needs a newer Starlette than the desktop app's API pins. Keep them separate.
🧩 Skills
Skills turn your clean library into finished work.
A Skill is a prompt-pack you run in the Claude desktop app or Codex, pointed at your library. The app never calls an LLM — nothing about Skills runs on anyone's servers. You bring your own AI; FilingForge gives it clean, cited source material and a precise prompt.
| Skill | What it does | |
|---|---|---|
| Business Model Brief | A cited, analyst-grade brief on how the company actually makes money — revenue mix, concentration, unit economics, moat — rendered as a clean, self-contained interactive HTML report by your own AI (a coding agent like Claude Code or Codex, which has file access). | Free |
| Concall Decoder | Every earnings call in your library, decoded — management's guidance track record (kept vs missed), tone shifts, what analysts keep asking, and what management avoids. A cited read on how much to trust the team. | ₹3,000 |
Skills are one-time purchases — you download a .md file and import it. No subscription, no licence server, no network check to run. The free Skill works identically to paid ones; the difference is what the prompt does.
Everything stays prompt-pack shaped: open, inspectable, and run on your machine with your own model.
🛠️ Build from source
Building locally also gives you a no-quarantine build on macOS.
If you only want the Python engine, start with docs/ENGINE.md.
Prerequisites
- Rust (stable toolchain) — for Tauri
- Node 20 — for the UI
- Python 3.11 — for the engine sidecar
1 — Python engine + sidecar
From the repo root, create a virtualenv and install the engine with its API + build extras:
python3.11 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -e ".[api,dev]" pyinstaller
Then build the engine into the Tauri sidecar binary:
python sidecar/build_sidecar.py
This produces ui/src-tauri/binaries/filingforge-api-<target-triple> (with a .exe suffix on Windows), which Tauri bundles as an external binary.
2 — UI dependencies
npm ci --prefix ui
3 — Build the desktop app
npm run tauri build --prefix ui
The bundled installers land under ui/src-tauri/target/release/bundle/ (.dmg on macOS, NSIS .exe on Windows).
Want a hot-reload dev loop instead? Run
npm run tauri dev --prefix ui(build the sidecar first so the app can spawn it).
🏗️ Tech
- Shell: Tauri v2 — a small, native desktop window on macOS and Windows.
- UI: React + TypeScript (Vite).
- Engine: a bundled Python FastAPI sidecar, frozen with PyInstaller, that does the downloading, PDF→Markdown conversion, and indexing.
- Wiring: the engine binds loopback only (
127.0.0.1:8765) and is spawned by the app as a sidecar process — never0.0.0.0, so it's not exposed to your LAN or the internet (only local processes on your own machine can reach it, which the Security policy treats as a threat surface).
🔒 Privacy / local-first
FilingForge is built to be boring about your data:
- No account. Nothing to sign up for.
- No telemetry. The app doesn't phone home.
- No LLM calls. FilingForge never sends your filings or queries to any model.
- Your files, your disk. Filings, Markdown, and indexes all live in a local library folder you control. The only outbound network traffic is fetching public filings from BSE and checking GitHub for app updates — never your data.
🤝 Contributing
Issues and pull requests are welcome. See CONTRIBUTING.md to get started, and please follow the Code of Conduct.
🛡️ Security
Found a vulnerability? Please report it responsibly — see SECURITY.md.
📄 License
MIT — see LICENSE. © 2026 Nandan Pugalia.
❤️ Support
FilingForge is free and open-source, built and maintained by one person at AlphaForge Integrated Solutions. If it saves you time, you can support its development via UPI from the in-app support screen. Thank you.
Yorumlar (0)
Yorum birakmak icin giris yap.
Yorum birakSonuc bulunamadi