User Flow

For the complete documentation index, see llms.txt. Markdown versions of documentation pages are available by appending .md to page URLs; this page is available as Markdown.

AI Chart Analysis — User Flow

From Prompt to Annotated Chart

A single analysis pass moves through the same sequence every time:

1. Open Analysis Mode on any chart.

2. Optionally select something — a candle, multiple candles, a price range, a swing high and low, an existing drawing, or just the currently visible area. Selection is optional; if you skip it, the AI reads the visible chart.

3. Enter your instruction — a natural-language sentence, a recognized keyword, or both together.

4. The AI identifies the requested analysis type(s). If you didn't name one, it selects the most applicable common analysis for the current chart and your preferences (see Analysis Types).

5. The AI returns a structured, versioned analysis plan — never raw numbers, never chart coordinates (see Data Contracts).

6. The deterministic engine executes the plan against OHLCV and available market data.

7. The engine returns, all with stable object IDs:

   • Evaluations with discrete confirmation states

   • Measurements and indicator values

   • Market structures

   • Drawing objects with exact chart coordinates

   • Warnings and invalidation conditions

8. The chart animates the result into place, layer by layer, so you can see how the analysis was built (see Chart Behavior).

9. A brief summary appears beside or below the chart.

10. You refine, edit, remove, lock, save, or rerun the analysis as needed (see Editing & Reuse).

The Closed Loop

The flow above isn't a one-way pipeline that starts over on every request. The engine's results — object IDs, detected structures, and their states — are retained as your active analysis context and fed back to the AI planner on every refinement. That's what makes a follow-up instruction like "use the previous swing low" work: the planner can see the structures the engine already found and bind directly to one of them by ID, instead of guessing at coordinates from scratch.

        Your Prompt + Chart Context + Preferences
                          ↓
        ┌───────────────────────────────────────┐
        │             AI Analysis Planner        │◄──────────────┐
        └───────────────────────────────────────┘               │
                          ↓                                      │
          Structured, Versioned JSON Analysis Plan               │
                          ↓                                      │
        ┌───────────────────────────────────────┐               │
        │      Deterministic Analysis Engine     │               │
        └───────────────────────────────────────┘               │
                          ↓                                      │
   Evaluations + Measurements + Indicators + Drawings            │
        (all with stable object IDs and states)                  │
                          ↓                                      │
             OHLCV Chart Visualization Layer                     │
                          ↓                                      │
        Brief Summary + Editable Chart Analysis                  │
                          ↓                                      │
     Active Analysis Context (IDs, states, your edits) ──────────┘
              fed back on every refinement

Every refinement is planned against this active context, not against a blank chart — so "remove the Fibonacci," "use the earlier swing low," or "keep the range and drop everything else" all resolve against structures the engine already identified, rather than requiring you to re-describe the whole analysis from the beginning.

Next Steps

• See Analysis Types for what you can ask for and how automatic selection works.

• See Editing & Reuse for the full refinement, save, and rerun workflow.