penecho/penecho

PenEcho – AI‑augmented visual canvas

What it is – PenEcho is an open‑source desktop/web application that turns a huge, infinite‑size drawing surface (20 000 × 20 000 px) into an interactive workspace where you can write, sketch, paste documents, and ask a large‑language model to generate explanations, formulas, plots, diagrams, or other visual widgets right next to your ink. The AI works as an “agent” that can read files, browse the web, and iteratively refine visual output without leaving the canvas.

Core capabilities

• PenEcho Agent – a multi‑step workflow that can ingest PDFs, Word, PowerPoint, Excel, images, code, or whole read‑only folders, combine them with web search and the current canvas, then produce and edit visual drafts.
• Visual Explorer – converts dense research material into a responsive, layered visual map with overview, detail, and evidence panels, reducing copy‑paste and tool‑switching.
• Spatial AI interaction – you write a question, equation, or diagram anywhere on the canvas; the model reads the surrounding marks and replies beside them.
• Editable AI widgets – generated HTML, professional diagrams, animations, or live‑data plugins appear as movable objects that can be refined in‑place via incremental diff edits.
• Multiple model back‑ends – supports up to ten connections (Kimi, Claude, OpenAI, DeepSeek, GLM, etc.) selectable per request; configuration is handled through penecho configure or a simple env file.
• PenEcho Cloud (optional) – sync private projects across devices, share public “Crafts” (called Echoes), and obtain a relay for remote LAN access while keeping API keys on the host.
• Security‑first design – API keys never leave the server process; a six‑digit LAN code guards each session; all data stays on the user’s machine unless they opt‑in to Cloud sync.

How it works

1. The browser sends only the cropped region of ink that the user selects, together with geometry metadata.
2. The local Node.js server validates the request, forwards it to the chosen executor (API call or local CLI), and receives a structured draft.
3. The draft is returned as a movable widget; the user can accept, discard, or edit it before it becomes permanent ink.
4. The canvas stores ink as sparse 512 × 512 tiles, so even a gigantic logical canvas remains lightweight.

Getting started

• Desktop binary – download the latest release from the GitHub Releases page and run it.
• npm install – npm install -g penecho, then run penecho configure to pick a model source (API key or an authenticated Kimi/Codex/Claude CLI). Launch with penecho and open http://localhost:3888.
• From source – git clone https://github.com/penecho/penecho.git && npm install && npm start.

Model recommendations (from the README) – Claude Opus 4.8/5.0, Kimi K3, the gpt‑5.6 family, DeepSeek‑v4‑flash‑vision‑exp, and GLM‑5.3‑flash are tuned for the canvas workload. The table in the README lists effort levels (low/medium/high/xhigh) and typical token usage.

License – AGPL‑v3 (with an optional commercial license for closed‑source redistribution). All code, assets, and the name/logo are covered by the project’s trademark and contributor agreements.

Who it’s for – researchers, educators, engineers, or anyone who wants to blend free‑hand note‑taking with AI‑generated visual content without constantly switching between a chat window and a drawing tool.