Charts and AI
Chart summary
written from detected facts
Illustration only. The summary describes this chart. It does not forecast, and it is not a recommendation.
Choosing a library
| Library | Licence | Best for | Watch out for |
|---|---|---|---|
| TradingView Advanced Charts and Trading Platform | Licensed by TradingView on application, with conditions on how and where it is used | Brokers and platforms that want the chart most traders already know, with trading from the chart on the Trading Platform edition | You supply the datafeed and the broker adapter. Licence conditions and customisation limits need checking early. |
| Lightweight Charts | Open source (Apache 2.0), from TradingView, with an attribution requirement | Fast, small charts inside web and mobile apps, and custom products where you control every pixel | Drawing tools, most indicators and workspace features are yours to build. |
| SciChart | Commercial licence | Very large or very fast datasets: tick charts, order book heatmaps, many charts on one screen | Licence cost per developer, and a smaller pool of developers who know it well. |
| Highcharts Stock | Commercial licence | Reporting and dashboards where financial charts sit alongside other chart types | Less suited to tick-level data at high update rates. |
| Custom (Canvas or WebGL) | Yours | Visualisations no library does well, such as footprint charts, volume profile or depth heatmaps | You own the maintenance. Only worth it when a library genuinely does not fit. |
Where the custom row comes from
Our founder built the charting engine for a global investment bank’s trade surveillance desk: hundreds of millions of ticks a day per view, market depth and time series across venues on one screen, and zoom down to the nanosecond. That is the ceiling. Most trading products need a small fraction of it, which is why we will usually recommend a library first.
Reviewed October 2026. Licence terms change, so check current terms with each vendor. Product names are trademarks of their owners. Listing them here does not imply a partnership or an endorsement.
How we build it
Candles that match your platform: same session times, same time zone, same handling of gaps. Most bad chart analysis is bad data.
Code finds levels, ranges, swings, patterns and regimes. Each detector has written rules and is tested against labelled charts.
Detectors emit structured facts: a range between two prices, how long it has lasted, how often each edge was tested. Nothing about what happens next.
A model turns the facts into a short summary in approved wording. It cannot add facts it was not given, and it has no access to orders or accounts.
A filter rejects any summary with advice-like phrasing, price targets, predictions or guarantees. Rejected output falls back to a plain template.
Every summary is stored with its input facts, prompt version and model version, so compliance can see exactly what a client saw.
Descriptive, not advisory
Allowed: describes the chart
Blocked: suggests a trade or a future
Price has closed above its 20-day high for the first time since 4 March.
Price is breaking out. Consider buying.
The range has narrowed to its smallest width in 120 bars.
A big move is coming.
A head and shoulders structure matches our detection rules on the daily chart.
Head and shoulders confirmed. Target 1.0650.
Volume on this bar is 2.1 times its 20-bar average.
Smart money is accumulating here.
The 50-day average crossed below the 200-day average on 12 September.
A death cross means it is time to sell.
What you get
From the Playbook
Robinhood is rolling out AI trading agents and the Bank of England fears they will trade in lockstep. A herding detector, stress circuit and charts, in code.
17 min read
Robinhood has 150,000 agentic accounts and 30 million daily agent tool calls, with approval switchable off. The retail agent platform architecture, in code.
17 min read
CNBC found half of a Kalshi market's volume in $5,500 trades. Surveillance for size clustering and volume-vs-depth, plus probability charts with expiry.
16 min read
Questions
It can, but wording and presentation matter. UK rules distinguish between a recommendation to buy or sell a particular investment and factual information that applies to everyone. We build to the descriptive side of that line, and your compliance team should review the final wording before launch.
Where it helps. Rules-based detection is easier to explain and test, so it is our starting point. We add learned models for things rules handle badly, such as regime classification, and evaluate them against labelled data before they go live.
Whichever suits the job and your data policy. We have built with Claude and OpenAI models, and we design the language layer so the model can be swapped without rewriting the product.
Yes. Annotations, shapes and study overlays can be drawn onto TradingView's libraries and Lightweight Charts from detected facts, so clients see the analysis on the chart they already use.
It depends on how many instruments, timeframes and detectors you need, and on the state of your data. Scoping gives you a firm plan. A first detector set with summaries on one asset class is a sensible starting release.
Share a screen on a 30-minute call and we will tell you what we would change first, and what it would take.
