Bring your own agent

Use your own model with the chart renderer, without managed model inference.

Separate the model from the environment

Your assistant can see chart images, reason in its own context, and request new images with its selected drawing commands. Rendering does not invoke our language model. Managed analysis is a separate endpoint that intentionally calls the configured model.

An external-agent workflow

1. POST /v1/charts/renders  → chart image + data provenance
2. Give the returned PNG to your own vision-capable model
3. Let that model choose annotations and produce typed commands
4. POST /v1/charts/renders with those commands → annotated PNG
5. Save the image, objects and thesis in your own bot or application

Drawing commands

Example render commands

{
  "market": {
    "symbol": "BINANCE:BTCUSDT",
    "interval": "240",
    "range": {
      "from": "2023-01-01T00:00:00Z",
      "to": "2023-02-01T00:00:00Z"
    }
  },
  "view": {
    "width": 1280,
    "height": 720,
    "theme": "dark",
    "includeVolume": true
  },
  "commands": [
    {
      "type": "draw_horizontal",
      "price": 22000,
      "role": "support",
      "label": "Support to assess"
    }
  ]
}

Image jobs support the existing drawing vocabulary: lines, zones, fibs, trends, text, markers, paths and chart scenes. The shared renderer rejects unsupported commands instead of pretending they applied. Image jobs cannot navigate to a different symbol; create a new request for that market.

Talk directly to the chart’s agent

Use managed analysis when you want Agent Charting’s brain to do the work. Continue the returned run with a message, or request respond_only for a question that should not change the chart.

Managed analysis MCP tools