He open-sourced a $30k AI trading terminal, 21k stars on github

0:00Bloomberg charges about $30,000 a year

0:02for a black and orange trading terminal.

0:05There's a GitHub repo with 21,000 stars

0:07that says it can replace it for free.

0:10Let's open the codebase and find out

0:12what's actually in there. Quick context

0:14for anyone who hasn't worked next to

0:16one. A Bloomberg terminal is the trading

0:18workstation you'll see on every hedge

0:20fund and bank desk. Real-time prices,

0:23news, and the chat network every dealer

0:25is on. Around 30 grand per seat per

0:28year. The pitch for Fin

0:31that you get the same workflow for

0:33nothing. Big claim. Let's see what's

0:35actually in there. First surprise,

0:38GitHub tags this repo as Python. So

0:41you'd expect a script, a notebook, maybe

0:43a TY. It's not.

0:46The current version is a native C++

0:48desktop app built on QT for the UI with

0:52Python embedded inside it for the

0:53financial math. Window, menus,

0:56installers for Windows, macOS, and

0:59Linux. Which also means setup isn't pip

1:02install. It's either a platform

1:04installer or a build from source dance

1:07with C make, QT, and pinned compiler

1:10versions. And the repo itself is nearly

1:13800 megabytes. The marketing site

1:15advertises 19,000 instruments across

1:18five exchanges. Where's that coming

1:20from? The repo wires up more than 100

1:23data connectors. The recognizable ones,

1:26Yahoo Finance, Polygon, Kraken, FRED,

1:30IMF, World Bank, DBnomics, AKshare, plus

1:36a long tail of government APIs. Notice

1:39what these have in common. They're the

1:41free public feeds. The same data a

1:43Pandas notebook can already pull. What's

1:46missing is what Bloomberg actually

1:48charges for.

1:49Curated proprietary feeds. The level two

1:52order book and the dealer chat network.

1:55So Fin Sept is a genuinely nice

1:58aggregator.

2:00It's not a data mode. The broker layer

2:02is where it gets interesting. 16

2:04integrations for live order placement.

2:06Four of them you'll recognize globally:

2:09Interactive Brokers, Alpaca, Tradier,

2:11Saxo.

2:12The other 12 are Zerodha, Angel One,

2:16Upstox, Groww, 5Paisa, and basically

2:19every other major Indian retail broker.

2:22So, the marketing copy says hedge funds.

2:26The code says Indian retail trader with

2:29a Zerodha account.

2:31Both audiences can be real,

2:34but it's worth knowing who's actually

2:36shipping the pull requests. Here's the

2:38part I didn't expect: a folder called

2:40src/mcp,

2:43and it's not a stub. Inside: Max P C

2:46client, Max P provider, Max P service,

2:49Max P manager, Max P marketplace,

2:53terminal MCP bridge, tool retriever,

2:55schema validator. About 140 kilobytes of

2:59C++ wiring model context protocol

3:01servers straight into the terminal,

3:03which means you can plug external tools,

3:05a database, a scraper, a custom API into

3:09Fincept's agents the same way you would

3:11with Claude code or cursor. For an open

3:13source finance app, that's genuinely

3:15rare. The AI layer ships 37 agents out

3:19of the box. The investor ones are named

3:21after Buffett, Graham, Lynch, Munger,

3:24Klarman, and Marks.

3:26To be clear, these aren't trained on

3:28those people. They're LLM persona

3:30prompts with different system

3:32instructions, useful as a research aid,

3:35not oracles. The wiring underneath is

3:37the part that's actually impressive.

3:40Multi-provider out of the box, OpenAI,

3:42Anthropic, Gemini, Grok, Deep Seek,

3:45MiniMax, Open Router, and Ollama for

3:47local models. So, you can run a

3:49Buffett-style agent on Claude or one

3:52entirely offline through Ollama.

3:54That part is real and it works. So, does

3:56it actually run?

3:58Yes, it does. 54 screen modules,

4:01dashboard, equity research, derivatives,

4:04crypto, sensible, then it keeps going,

4:07maritime tracking, satellite data,

4:10geopolitical relationship maps, a

4:12polymarket screen, a node editor for

4:15visual workflows. Yes, this finance

4:17terminal also tracks ships and bets on

4:19prediction markets. The readme tagline

4:22says, "Your thinking is the only limit."

4:24The repo took that very seriously.

4:27Reality check.

4:29Four things worth sitting with. One,

4:3121,000 stars, 3,000 forks, last commit 2

4:36days ago.

4:37This is genuinely active, not a

4:39graveyard.

4:41Two, the marketing comparison to

4:43Bloomberg at 27 grand quietly skips over

4:46the data, the chat network, and the SLA

4:49grade reliability you're actually paying

4:51for. A free UI is not the same as a real

4:55moat.

4:56Three, the license is AGPL 3,

4:58dual-licensed. Fork this into a paid

5:01SaaS and you either open-source your

5:03fork or buy a commercial license.

5:06Open-source doesn't automatically mean

5:07free for your startup. Four, 21 months

5:11old, already on version four.

5:13That velocity is either heroic or

5:16AI-assisted. Realistically, a bit of

5:18both and that's worth pricing into the

5:20maintenance risk. So,

5:23is it a Bloomberg killer?

5:25No. Not even close and the comparison

5:28was never quite fair.

5:30But, it is, I'd argue, the most

5:32ambitious free finance workstation

5:34open-source has produced and the MCP

5:37layer alone makes it worth a clone.

5:40Install it if you trade Indian retail,

5:42if you want a real MCP-enabled quant

5:44playground, or if you just enjoy poking

5:47around ambitious OSS codebases. Skip it

5:50if you actually need Bloomberg-grade

5:52data, if you wanted a one-line pip

5:54install, or if AGPL gives you legal

5:56heartburn. If you'd like more repo

5:59audits like this one, the two or three

6:01files that actually make a project

6:02interesting, consider subscribing. See

6:05you in the next one.