Transcription
Bloomberg charges about $30,000 a year for a black and orange trading terminal. There's a GitHub repo with 21,000 stars that says it can replace it for free. Let's open the codebase and find out what's actually in there.
Quick context for anyone who hasn't worked next to one. A Bloomberg terminal is the trading workstation you'll see on every hedge fund and bank desk. Real-time prices, news, and the chat network every dealer is on. Around 30 grand per seat per year. The pitch for Fin that you get the same workflow for nothing. Big claim. Let's see what's actually in there.
First surprise, GitHub tags this repo as Python. So you'd expect a script, a notebook, maybe a TY. It's not. The current version is a native C++ desktop app built on QT for the UI with Python embedded inside it for the financial math. Window, menus, installers for Windows, macOS, and Linux. Which also means setup isn't pip install. It's either a platform installer or a build from source dance with C make, QT, and pinned compiler versions. And the repo itself is nearly 800 megabytes.
The marketing site advertises 19,000 instruments across five exchanges. Where's that coming from? The repo wires up more than 100 data connectors. The recognizable ones, Yahoo Finance, Polygon, Kraken, FRED, IMF, World Bank, DBnomics, AKshare, plus a long tail of government APIs. Notice what these have in common. They're the free public feeds. The same data a Pandas notebook can already pull. What's missing is what Bloomberg actually charges for. Curated proprietary feeds. The level two order book and the dealer chat network. So Fin Sept is a genuinely nice aggregator. It's not a data mode.
The broker layer is where it gets interesting. 16 integrations for live order placement. Four of them you'll recognize globally: Interactive Brokers, Alpaca, Tradier, Saxo. The other 12 are Zerodha, Angel One, Upstox, Groww, 5Paisa, and basically every other major Indian retail broker. So, the marketing copy says hedge funds. The code says Indian retail trader with a Zerodha account. Both audiences can be real, but it's worth knowing who's actually shipping the pull requests.
Here's the part I didn't expect: a folder called src/mcp, and it's not a stub. Inside: Max P C client, Max P provider, Max P service, Max P manager, Max P marketplace, terminal MCP bridge, tool retriever, schema validator. About 140 kilobytes of C++ wiring model context protocol servers straight into the terminal, which means you can plug external tools, a database, a scraper, a custom API into Fincept's agents the same way you would with Claude code or cursor. For an open source finance app, that's genuinely rare.
The AI layer ships 37 agents out of the box. The investor ones are named after Buffett, Graham, Lynch, Munger, Klarman, and Marks. To be clear, these aren't trained on those people. They're LLM persona prompts with different system instructions, useful as a research aid, not oracles. The wiring underneath is the part that's actually impressive. Multi-provider out of the box, OpenAI, Anthropic, Gemini, Grok, Deep Seek, MiniMax, Open Router, and Ollama for local models. So, you can run a Buffett-style agent on Claude or one entirely offline through Ollama. That part is real and it works.
So, does it actually run? Yes, it does. 54 screen modules, dashboard, equity research, derivatives, crypto, sensible, then it keeps going, maritime tracking, satellite data, geopolitical relationship maps, a polymarket screen, a node editor for visual workflows. Yes, this finance terminal also tracks ships and bets on prediction markets. The readme tagline says, "Your thinking is the only limit." The repo took that very seriously.
Reality check. Four things worth sitting with.
One, 21,000 stars, 3,000 forks, last commit 2 days ago. This is genuinely active, not a graveyard.
Two, the marketing comparison to Bloomberg at 27 grand quietly skips over the data, the chat network, and the SLA grade reliability you're actually paying for. A free UI is not the same as a real moat.
Three, the license is AGPL 3, dual-licensed. Fork this into a paid SaaS and you either open-source your fork or buy a commercial license. Open-source doesn't automatically mean free for your startup.
Four, 21 months old, already on version four. That velocity is either heroic or AI-assisted. Realistically, a bit of both and that's worth pricing into the maintenance risk.
So, is it a Bloomberg killer? No. Not even close and the comparison was never quite fair. But, it is, I'd argue, the most ambitious free finance workstation open-source has produced and the MCP layer alone makes it worth a clone. Install it if you trade Indian retail, if you want a real MCP-enabled quant playground, or if you just enjoy poking around ambitious OSS codebases. Skip it if you actually need Bloomberg-grade data, if you wanted a one-line pip install, or if AGPL gives you legal heartburn.
If you'd like more repo audits like this one, the two or three files that actually make a project interesting, consider subscribing. See you in the next one.