# OpenChart lets AI agents work inside TradingView charts > Source: > Published: 2026-10-06 16:29:05+00:00 # OpenChart lets AI agents work inside TradingView charts OpenChart is an open‑source alternative to TradingView that lets you bring AI agents directly onto your charts. Instead of typing commands in a terminal, you can draw shapes, annotate setups, write indicators or create alerts using natural language through agents like [Claude](https://promptcube3.com/en/tags/claude/) or Codex. The agent receives the chart context, can run custom indicators written in the Tea language, and saves any research or notes back into the same workspace where the chart lives. Alerts are triggered when price crosses a user‑drawn shape; at that moment the agent investigates the move and stores the findings alongside the chart for later review. All data stays on your machine. The application, its backend and your chart files are stored locally, which means nothing leaves your computer unless you opt for the paid cloud market data feed that promises lower latency. Free market data is available from day one, so you can start experimenting without any subscription. The project is hosted on GitHub and the source code is publicly available for anyone to inspect, fork or contribute to. Right now the distributed binary runs only on macOS with Apple Silicon. If you try to launch the app on an Intel‑based Mac or on Linux/Windows you will encounter a compatibility error and the program will not start. In that case the next step is to visit the repository’s “Building from source” instructions, check the issue tracker for any community‑made cross‑platform builds, or consider running the software in a virtualized macOS environment until official support expands. The developers have stated they are working on broader platform coverage, so watching the releases page for updates is a sensible follow‑up. Custom logic is expressed in Tea, a domain‑specific language created by the OpenChart team for market‑related scripting. You can write an indicator that calculates a moving average crossover, attach it to a chart, and then ask your agent to tweak the parameters based on recent price action. Because the indicator code lives alongside the chart, you can version‑control it with the rest of your research notebook, making it easy to share or revisit later. The workflow is deliberately frictionless: open a chart, draw a trendline or support‑resistance zone, attach an alert, and let the agent handle the rest. When the alert fires you receive a notification that the agent has logged its analysis, which you can open directly from the workspace to see the chain of thought, any generated code, and the resulting notes. This tight coupling reduces the back‑and‑forth between a separate IDE and your charting tool, keeping the focus on the market move itself. If you are interested in experimenting with AI‑driven chart analysis, the easiest entry point is to download the macOS arm64 DMG from the releases page, install it, and connect your preferred agent via the provided API key fields. Once the agent is linked, try creating a simple alert on a horizontal line and observe how the agent investigates the breach. Should you hit any roadblocks, the GitHub issues section is active and the maintainers respond to questions about build errors, missing dependencies, or feature requests. Overall, OpenChart offers a way to combine the visual intuition of traditional charting with the programmable power of LLM agents, all while keeping your data private and your workflow contained within a single application. The project is still early, but the core concepts—local storage, agent‑chart interaction, and the Tea scripting language—are already functional and ready for community testing. [Next Self-bench lets you benchmark coding agents using your own PRs →](https://promptcube3.com/en/threads/9817/) ## All Replies （1） Want a live back-and-forth? [Join the global AI chat room](https://promptcube3.com/en/chat/) — login to talk. Tea language compiles to WASM so indicators run at native speed — backtesting still needs a separate engine though.