spovalova/openterminal

Open Terminal is worth a look if you want a local, question-driven research surface over market data and are comfortable that it is an early, deliberately narrow tool. It is not a trading terminal or an execution venue.

The problem

Researching a stock usually means opening a quote page, a chart, filings, news and an option chain in separate tabs, then manually stitching together an answer. Open Terminal changes the first step: you type a question such as whether a drop is news-driven or market-wide, and the agent produces four to six live widgets plus a short reply. The workspace is rebuilt on each follow-up, so the answer and the evidence stay in one place instead of being reassembled by hand. The trade-off is that you are researching through an LLM's selected queries rather than browsing freely.

What's different

The agent has a small tool set: search a catalog of data routes, run a recipe or a single route, transform results with pandas, and replace the workspace. Crucially, the reply is constrained: the agent cannot type a price, its figures must come from frame previews it saw, and the server checks those numbers against the frames. Each claim links to the widget that displays it, which is the part that makes an AI answer usable for financial work.

Data comes through connectors. Three ship in the box, including OpenBB's 834 routes answered in process by TypeScript ports of its providers, with Yahoo and Finnhub alongside; a recipe can name fallbacks so OpenBB is tried before Yahoo. A running OpenBB API adds routes the ports do not cover, and the connector surface is a single defineConnector function with descriptors and a call handler, so adding a source is a contained piece of work rather than a fork.

That design has a real cost. Widgets poll on a per-route TTL rather than stream, and the roadmap explicitly leaves out diff or accept/reject of agent edits, drag-and-resize, alerts, broker connections, crypto and non-US tickers. It is also key-dependent for the conversational part: without an Anthropic or OpenAI key you get a sample workspace, not an agent.

Who should try it

A good fit is a technically comfortable investor or analyst who already has an Anthropic or OpenAI key, wants to explore US equities and options from a local machine, and would rather extend a connector than wait for a vendor to add a source. The documented no-key data providers make it useful for basic quotes, charts, filings and macro even before the agent is configured.

A poor fit is anyone who needs order execution, real-time streaming, crypto or non-US coverage, or a polished, drag-and-drop charting environment; all of those are explicitly out of scope in the current roadmap. Someone who will not run an LLM against market data, or who needs signed installers, should also look elsewhere.

Requirements and limitations

The desktop app for macOS, Windows and Linux carries its own runtime, the server and a Python runtime (Pyodide with pandas, numpy and scipy), so there is nothing else to install; builds are unsigned, and the install script is the documented way to avoid the browser-download warnings. Running from source requires Node 22 and pnpm, and the browser path via npx openterminal is documented but the package is not yet on npm, so from-source use is the current fallback. The agent needs an Anthropic or OpenAI key; the first launch asks for one, and skipping it leaves you with a sample workspace. Data quality and coverage therefore depend on the connectors you configure, and the agent's caching and per-route TTL mean freshness varies by route. Ongoing upkeep means watching the unsigned, self-updating release channel and keeping your own connector descriptors current.

HotRepo's take

HotRepo has only a single dated observation: 235 total stars and #23 on the NEW-REPOSITORY board for TypeScript on 2026-10-25, with no daily-board history to compare. With a six-day-old repository and one snapshot, persistence cannot yet be judged. Before investing time, check the connector descriptor format and whether the routes you need are among those answered in process rather than requiring a live OpenBB API.

Project source and documentation