PenEcho Releases Handwritten Math AI Canvas

PenEcho is an open-source project that eliminates the hassle of converting handwritten formulas or shapes on paper into text by having AI instantly recognize pen-drawn content on a canvas and draw the answers alongside it.

Key Technical Features

• Efficient Canvas Processing: In a 20,000 x 20,000 logical coordinate system, only inked areas are split into 512 x 512 tiles for processing, minimizing browser memory load.
• Secure Execution Structure: The server validates and processes only the images and coordinates around the handwriting received from the client, ensuring security by not directly executing JavaScript returned by the model.
• Draft Layer System: Answers generated by AI are displayed on a separate draft layer that does not mix with confirmed handwriting, and are saved as confirmed objects only after user approval.

Model Integration and Costs

• Various LLMs can be integrated via Claude CLI, Codex CLI, or OpenAI/Anthropic-compatible APIs.
• Recommended Models: claude-opus-4-8 (for daily tasks), gpt-5.6-sol (speed priority), and Fable 5 (quality/speed balance) are recommended.
• Costs by Reasoning Level: Low level consumes approximately 1,000 output tokens per request, while High/Xhigh levels consume 5,000–8,000 tokens, with costs varying by model unit price.

Installation and Security

• Installable via the npm install -g penecho command, with the default port being 3888.
• Network access requires authentication with a 6-digit security code, and users are advised to avoid exposing the tool to the public internet when integrating with command-line tools.