# ChatCanvas for DeepSeek Harness

[English](README.md) | [简体中文](README.zh-CN.md)

[Download preview](https://github.com/Tryanstm/dsh-chatcanvas/releases) · [Installation](docs/INSTALL.md) · [Report an issue](https://github.com/Tryanstm/dsh-chatcanvas/issues)

**Turn AI conversations into a workspace you can edit, explore and play.**

ChatCanvas is a third-party plugin for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness). It brings rich messages, editable canvases, interactive UI previews, math plots and music editing into the conversation.

## See it in action

Three short demonstrations, edited from real recordings. Play the videos below, or download their MP4 files. Waiting and typing are accelerated; the clips stop before subsequent AI replies. The interface shown is Simplified Chinese, with bilingual video captions.

### Write clearly · 21 seconds

Headings, quotes and colors; refine math notation and add a short instruction. The opening formatting example is a still screenshot, labeled in the video.

https://github.com/user-attachments/assets/b8679eac-9fc5-479a-9335-aa826f832c0c

[Download MP4](https://github.com/Tryanstm/dsh-chatcanvas/raw/refs/heads/main/media/demos/compose.mp4)

### Sketch your idea · 26.5 seconds

Insert a canvas and draw a layout directly in the composer, including a graph area, controls and annotations.

https://github.com/user-attachments/assets/82bdbab1-b996-42c0-a54e-44af016d4413

[Download MP4](https://github.com/Tryanstm/dsh-chatcanvas/raw/refs/heads/main/media/demos/canvas.mp4)

### Edit the result · 24 seconds

Select and drag controls, adjust a result panel and reposition text in a UI draft.

https://github.com/user-attachments/assets/5737babc-d01e-439c-95f1-6cfb779e6cde

[Download MP4](https://github.com/Tryanstm/dsh-chatcanvas/raw/refs/heads/main/media/demos/ui-editor.mp4)

[About these recordings and adding more demos](docs/DEMOS.md)

## What you can do

| Feature | In the conversation |
| --- | --- |
| Rich input and messages | Writing boxes, plain-text boxes, code highlighting, copying and native file references |
| Canvas | Draw, erase, arrange objects, snap to guides, control layers and share editable copies |
| Interactive UI | Build pages and controls, edit visual programs and try the result locally |
| Math | Explore functions, equations, parameters and 2D/3D plots |
| Music | Edit notes and tracks, audition instruments, import MIDI, export MIDI/WAV and transcribe WAV/MP3 locally |
| Visual questions | Let the AI show an editable draft and ask you to adjust it before proceeding |

<details>
<summary>Capability overview</summary>

![Capability overview — not an application screenshot](docs/assets/features.svg)

</details>

## Install

This is a **preview release**. Tested integration targets: DeepSeek Harness **0.2.0-rc.2 and 0.2.1-rc.2**. The installation helper is for Windows. Other host versions and operating systems have not been verified.

1. Open [Releases](https://github.com/Tryanstm/dsh-chatcanvas/releases) and download `local-dsh-chatcanvas-1.4.0-rc.12.tgz`.
2. Fully quit Harness.
3. Install through the desktop app's bundled CLI or its supported local-package installer.
4. Reopen Harness. The plugin settings section is **对话画布 · ChatCanvas**.

The GitHub-generated **Source code** archive is for developers. It is not the compiled installation package. See [installation and rollback](docs/INSTALL.md).

**npm / Discover:** the first `dsh-chatcanvas@1.4.0-rc.13` publication is being prepared. It is not yet available on npm, and the market source currently has no entries. Keep using the GitHub Release above until publication is verified. See [publishing, the market source and migration from the old local package](docs/PUBLICATION.md#english).

## Start using it

Write normally, or right-click the composer to insert content. Ask the AI to create a UI preview, math plot, music project or shared canvas. Use an editable copy when you want to modify an already-sent work.

Example prompts:

- “Create a two-page timer interface that I can try here.”
- “Plot y = a*x² and let me change a.”
- “Make a short piano melody and show the notes for editing.”
- “Show an editable layout draft and ask me to adjust it.”

The interface currently uses **Simplified Chinese**. This repository provides English and Chinese documentation. Full English UI localization is not included in this preview.

## Settings

Open **Settings → 对话画布 · ChatCanvas**. The public plugin exposes only two preferences:

| Preference | Default | How to change it |
| --- | --- | --- |
| Program recommendations / 程序推荐选项 | 2 | Click the count, choose 0–5, then save |
| AI recommendations / AI 推荐选项 | 1 | Click the count, choose 0–5, then save |

Execution auditing, audit switches and tool/Skill whitelists are not part of this plugin. The host's permission controls remain available. Existing document and storage identifiers are retained for compatibility with older works.

## Permissions, privacy and cost

ChatCanvas does not register an execution-review gate or change the host's permission presets. Host permissions continue to apply. Canvas/UI editing and audio transcription use local code and packaged assets. AI text recommendations, translations, their quality checks and automatic conversation titles use the model configured in Harness and may incur extra usage. See [privacy and model usage](docs/PRIVACY.md).

## Preview limitations

- Audio transcription is an editable estimate; dense mixes and overlapping instruments can produce incorrect notes.
- Sampled playback is not a professional instrument library. Some instruments sound more convincing than others.
- UI simulation checks program logic; it does not replace visual testing.
- Long drawing sessions, high-DPI rendering and interactions with other plugins need further real-window testing.
- Normal Markdown tables remain supported. The old standalone enhanced-table editor is retired.

## Development

See [build and test instructions](docs/DEVELOPMENT.md), [integration](docs/INTEGRATION.md) and [document protocol](docs/PROTOCOL.md).

## Feedback

Use the issue templates for bug reports and feature requests. Include the plugin version, Harness version, operating system and steps to reproduce. Share sanitized examples and screenshots. Avoid API keys and private conversations.

## License and credits

Plugin code: [MIT](LICENSE). Bundled assets retain their own licenses: TimGM6mb SoundFont is **GPL-2.0-only**; Basic Pitch and TensorFlow.js assets are **Apache-2.0**. See [third-party notices](THIRD_PARTY_NOTICES.md) and bundled license/source material. The MIT label for plugin code does not relicense third-party assets.

This project is not an official DeepSeek product and is not affiliated with or endorsed by DeepSeek. Upstream DeepSeek Harness components retain their original notices.