Skip to content

v1.21.0

Latest

Choose a tag to compare

@github-actions github-actions released this 20 Nov 21:21

AI agents and the AI Chat

AVASvgMaker can now be driven by an AI model, the same way as AVA Modeler, AVALightWave, CirqAvalonia and AVAPageStream.

  • AI Chat (F12, or AI › AI Chat). Ask "examine src/orders.py and flowchart it for me" and it reads the file, works out its control flow and draws the flowchart on a new page, laid out, then looks at it and tidies it. Or ask for any drawing: a network, a swimlane, an approval process. Every step is one Ctrl+Z, and Undo this reply takes back a whole answer; Timeline and History (if you opt in) keep track.
  • Any model. Claude with your Anthropic key; a model on your own computer or network through Ollama, LM Studio, llama.cpp or vLLM, with no key; or OpenAI and other cloud services with their key. Setup… chooses it.
  • MCP server with 32 tools and 3 recipes: pages, stencils, shapes, connectors, arranging and automatic layout, whole flowcharts in one call, Mermaid in and out, reading files in the working folder, a picture of any page, and export to SVG, PDF, PNG, JPEG, WebP, BMP, Mermaid and Visio. Batches as one undo step (or a dry run), checkpoints, and a session kept as a reusable steps file with parameters.
    • AVASvgMaker --mcp --attach: an agent such as Claude Code draws in the open window (after AI › Allow AI Agents).
    • AVASvgMaker --mcp: no window, a drawing of its own.
    • Over the network on port 8769 with an access token (AI › Connect an Agent).
  • AI › Working Folder… sets where the chat may read and write; nothing outside it can be read.

File format 20: every shape now carries a lasting id, so agents' references hold through undo, saving and opening. Older files still open; files saved by 1.21.0 need 1.21.0 or later.

Full changelog: v1.20.0...v1.21.0