kunthawat/moreminimore-vibe
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update the rebranding and fix issues

Browse Source
This commit is contained in:
Kunthawat Greethong
2025-12-22 10:14:05 +07:00
parent 29248688f3
commit 68189db3b3
917 changed files with 2371 additions and 138265 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
docs/architecture.md
View File

@@ -14,9 +14,9 @@ If you're not familiar with Electron apps, they are similar to a full-stack Java
The core workflow of Dyad is that a user sends a prompt to the AI which edits the code and is reflected in the preview. We'll break this down step-by-step.
1. **Constructing an LLM request** - the LLM request that Dyad sends consists of much more than the prompt (i.e. user input). It includes, by default, the entire codebase as well as a detailed [system prompt](https://github.com/dyad-sh/dyad/blob/main/src/prompts/system_prompt.ts) which gives the LLM instructions to respond in a specific XML-like format (e.g. `<dyad-write path="path/to/file.ts">console.log("hi")</dyad-write>`).
2. **Stream the LLM response to the UI** - It's important to provide visual feedback to the user otherwise they're waiting for several minutes without knowing what's happening so we stream the LLM response and show the LLM response. We have a specialized [Markdown parser](https://github.com/dyad-sh/dyad/blob/main/src/components/chat/DyadMarkdownParser.tsx) which parses these `<dyad-*>` tags like the `<dyad-write>` tag shown earlier, so we can display the LLM output in a nice UI rather than just printing out raw XML-like text.
3. **Process the LLM response** - Once the LLM response has finished, and the user has approved the changes, the [response processor](https://github.com/dyad-sh/dyad/blob/main/src/ipc/processors/response_processor.ts) in the main process applies these changes. Essentially each `<dyad-*>` tag described in the [system prompt](https://github.com/dyad-sh/dyad/blob/main/src/prompts/system_prompt.ts) maps to specific logic in the response processor, e.g. writing a file, deleting a file, adding a new NPM package, etc.
1. **Constructing an LLM request** - the LLM request that Dyad sends consists of much more than the prompt (i.e. user input). It includes, by default, the entire codebase as well as a detailed [system prompt](https://github.com/kunthawat/moreminimore-vibe/blob/main/src/prompts/system_prompt.ts) which gives the LLM instructions to respond in a specific XML-like format (e.g. `<dyad-write path="path/to/file.ts">console.log("hi")</dyad-write>`).
2. **Stream the LLM response to the UI** - It's important to provide visual feedback to the user otherwise they're waiting for several minutes without knowing what's happening so we stream the LLM response and show the LLM response. We have a specialized [Markdown parser](https://github.com/kunthawat/moreminimore-vibe/blob/main/src/components/chat/DyadMarkdownParser.tsx) which parses these `<dyad-*>` tags like the `<dyad-write>` tag shown earlier, so we can display the LLM output in a nice UI rather than just printing out raw XML-like text.
3. **Process the LLM response** - Once the LLM response has finished, and the user has approved the changes, the [response processor](https://github.com/kunthawat/moreminimore-vibe/blob/main/src/ipc/processors/response_processor.ts) in the main process applies these changes. Essentially each `<dyad-*>` tag described in the [system prompt](https://github.com/kunthawat/moreminimore-vibe/blob/main/src/prompts/system_prompt.ts) maps to specific logic in the response processor, e.g. writing a file, deleting a file, adding a new NPM package, etc.
To recap, Dyad essentially tells the LLM about a bunch of tools like writing files using the `<dyad-*>` tags, the renderer process displays these Dyad tags in a nice UI and the main process executes these Dyad tags to apply the changes.