This repository is an article-derived reference project based on the original MQL5 article. It does not claim to reproduce the full original source code unless files are explicitly attached.
This repository documents and references the MQL5 article **“Building AI-Powered Trading Systems in MQL5 (Part 12): Giving the Assistant Chart Vision and Tool Access.”** It is an article-derived MQL5 project focused on extending an AI trading assistant with two major capabilities:
- chart vision through screenshots and image attachments
- tool access for live terminal, market, account, and calendar data
The article describes how an Expert Advisor-based assistant can move beyond plain text chat by sending chart images to a multimodal model and by allowing the model to request specific live data through a controlled tool-calling workflow.
## Original Article
- **Article ID:** 23893
- **Title:** Building AI-Powered Trading Systems in MQL5 (Part 12): Giving the Assistant Chart Vision and Tool Access
This repository exists to preserve the structure and technical ideas presented in the original article in a reusable repository format.
A reader can use this repository to:
- understand how chart screenshots can be captured and sent to an AI model from MQL5
- study a multimodal message flow that combines text and image content
- learn how tool-calling can expose controlled terminal data to an assistant
- reuse the article’s architectural ideas for MQL5 dashboard panels, image rendering, and request orchestration
- inspect the project file layout explicitly listed in the article attachments
Because this repository is article-derived, its completeness depends on the files actually attached or otherwise provided by the author.
## Key Concepts
- Trading Systems
- Integration
- Expert Advisors
- Statistics
- Multimodal AI messaging
- Chart screenshot capture
- Base64 image encoding
- Tool calling
- Live market context injection
- Chart object inspection
- Trade history summarization
- Economic calendar access
- Canvas-based UI rendering
## Algorithm / Architecture Summary
The article presents a combined UI, image-processing, and AI-integration workflow for an MQL5 assistant.
1.**Capture chart imagery**
- The assistant captures the current chart as a screenshot.
- The panel UI is temporarily hidden so the screenshot contains only the chart.
- PNG is used for request payload/archive purposes, while BMP is used for local panel display.
2.**Encode the image for transport**
- The PNG bytes are converted to base64.
- Line breaks are removed so the encoded image can be embedded safely into JSON.
3.**Stage the attachment before sending**
- A pending attachment state stores:
- image base name
- width and height
- base64 payload
- The user can attach or remove the staged screenshot before sending the message.
4.**Build multimodal request content**
- The outgoing message becomes a content array rather than plain text.
- It includes:
- a text item for the prompt
- an image item using a base64 data URL
- This allows the model to reason about actual chart visuals instead of typed descriptions only.
5.**Render image previews inside the panel**
- The local BMP copy is decoded for display.
- Thumbnails are cached using a bounded cache strategy.
- Inline thumbnails and a compose-chip preview are drawn into the dashboard.
6.**Open a full-image viewer**
- Clicking a thumbnail opens a native-resolution viewer.
- The viewer supports:
- full-resolution image display
- vertical and horizontal scrolling
- grab-pan interaction
- close controls
- This is handled through dedicated viewer state and interaction routing.
7.**Clean up image files**
- Stored chat history references image markers.
- When a chat is deleted, related PNG/BMP files are also deleted.
8.**Define callable tools for the model**
- The article defines a JSON tool menu for the model with functions for:
- indicator retrieval
- open positions
- chart objects
- trade history
- economic calendar events
9.**Dispatch tool calls inside MQL5**
- Tool-call arguments are parsed from JSON.
- A dispatcher routes the request to the appropriate MQL5 function.
- Tool results are returned as readable text rather than raw binary or complex structures.
10.**Attach automatic market context**
- A live market snapshot can be appended automatically to each message.
- This context includes current symbol/timeframe details, prices, recent candles, positions, and account status.
11.**Run a tool-call loop during send**
- The assistant sends the request with optional vision content and tool definitions.
- If the model responds with tool calls:
- the terminal executes each requested tool
- tool outputs are appended to the conversation
- the request loop continues until the model returns final text or the round limit is reached
Overall, the architecture combines a chart-facing UI layer, an image handling pipeline, a tool exposure layer, and a controlled model interaction loop.
## Mentioned or Attached Files
### Explicitly attached files
The article explicitly lists these attached files:
-`AI Canvas Theme.mqh`
-`AI Canvas Primitives.mqh`
-`AI JSON FILE.mqh`
-`AI Canvas State.mqh`
-`AI Canvas Scrollbar.mqh`
-`AI Canvas Editor.mqh`
-`AI Canvas Render.mqh`
-`AI Logic.mqh`
-`AI Canvas Interact.mqh`
-`AI Canvas Shell.mqh`
-`AI EA PART 12.mq5`
-`MQL5.zip`
### ZIP attachment
An attached ZIP archive is available in the processed input:
The article also mentions internal image storage outputs used by the implementation:
- PNG screenshot copies
- BMP screenshot copies
## Statistics
- **Word count:** Not available
- **Reading time:** Not available
- **Image count:** Not available
- **Code block count:** Not available
- **File count:** Not available
## Tags
- mql5
- metatrader-5
- trading-systems
- integration
- expert-advisors
- statistics
- difficulty-advanced
## Difficulty
**Advanced**
Reason: the article combines MQL5 UI rendering, chart screenshot handling, base64 encoding, JSON request construction, multimodal AI messaging, tool schema design, live terminal data extraction, and iterative tool-call orchestration.
## Limitations
- This repository is derived from the article and may not reproduce the full original project unless the attached files are included here.
- The implementation depends on files actually attached or provided by the author.
- The README is based on article metadata and article content analysis, not on verification of every attached source file inside the ZIP archive.
- Internal ZIP contents were described in the article, but repository completeness still depends on what was actually extracted and committed.
- The project should not be assumed production-ready unless explicitly stated by the original author, which was not established in the processed input.
## Reference
- Original article: https://www.mql5.com/en/articles/23893
