vzcode 2.21.0 → 2.22.0

package/src/llm-streaming-server/README.md

@@ -0,0 +1,71 @@
+# LLM Streaming Server
+
+Server-side library for streaming LLM responses with ShareDB integration.
+
+## Overview
+
+This module is being extracted from VZCode's AI chat functionality to become a standalone, reusable library. It provides server-side functionality for:
+
+- **LLM Streaming**: Creating and managing streaming LLM functions with OpenAI/OpenRouter
+- **AI Code Editing**: Performing AI-assisted code transformations and edits
+- **ShareDB Integration**: Managing chat operations and real-time collaboration
+- **Request Validation**: Validating incoming AI chat requests
+- **Error Handling**: Handling errors gracefully in streaming contexts
+
+## Current Status
+
+⚠️ **Work in Progress**: This is Phase 1 of the extraction plan. The files have been copied to this directory and imports have been updated, but the code is not yet decoupled from VZCode.
+
+## Files
+
+- `index.ts` - Main entry point exporting all public APIs
+- `llmStreaming.ts` - Core LLM streaming functionality with OpenAI/OpenRouter
+- `aiEditing.ts` - AI-assisted code editing operations
+- `chatOperations.ts` - ShareDB chat operations (messages, events, streaming)
+- `validation.ts` - Request validation utilities
+- `errorHandling.ts` - Error handling for streaming contexts
+
+## Dependencies
+
+Currently depends on:
+
+- `openai` - OpenAI SDK for LLM interactions
+- `sharedb` - Real-time collaborative editing framework
+- `llm-code-format` - Markdown parsing for code blocks
+- `editcodewithai` - Code editing utilities
+- VZCode utilities: `../types.js`, `../ot.js`, `../randomId.js`, etc.
+
+## Future Plans
+
+This module will eventually be:
+
+1. Decoupled from VZCode-specific types and utilities
+2. Published as `@vizhub/llm-streaming-server` on npm
+3. Made framework-agnostic (can work with Express, Fastify, etc.)
+4. Fully documented with usage examples
+
+See the [main issue](https://github.com/vizhub-core/vzcode/issues/XXX) for the full extraction plan.
+
+## Usage (Current)
+
+Currently, this module is imported by `src/server/aiChatHandler/index.ts`:
+
+```typescript
+import { createLLMFunction } from '../../llm-streaming-server/llmStreaming.js';
+import { performAIEditing } from '../../llm-streaming-server/aiEditing.js';
+import {
+  ensureChatsExist,
+  ensureChatExists,
+  addUserMessage,
+  setAIStatus,
+} from '../../llm-streaming-server/chatOperations.js';
+```
+
+## Contributing
+
+As this module is being extracted, please:
+
+- Keep changes minimal and focused
+- Maintain backward compatibility with VZCode
+- Update tests when modifying functionality
+- Document any new dependencies or breaking changes

package/src/llm-streaming-server/aiEditing.ts

@@ -0,0 +1,102 @@
+import {
+  assembleFullPrompt,
+  prepareFilesForPrompt,
+} from 'editcodewithai';
+import { formatMarkdownFiles } from 'llm-code-format';
+import {
+  createFilesSnapshot,
+  generateFilesUnifiedDiff,
+} from '../utils/fileDiff.js';
+
+// Dev flag for waiting 1 second before starting the LLM function.
+// Useful for debugging and testing purposes, e.g. checking the typing indicator.
+const delayStart = false;
+
+/**
+ * Performs AI chat without editing - just generates a response
+ */
+export const performAIChat = async ({
+  prompt,
+  shareDBDoc,
+  llmFunction,
+}) => {
+  const { files } = prepareFilesForPrompt(
+    shareDBDoc.data.files,
+  );
+  const filesContext = formatMarkdownFiles(files);
+
+  // 2. Assemble the final prompt for Q&A mode
+  const fullPrompt = assembleFullPrompt({
+    filesContext,
+    prompt,
+    editFormat: 'whole',
+  });
+
+  if (delayStart) {
+    await new Promise((resolve) =>
+      setTimeout(resolve, 1000),
+    );
+  }
+
+  // Call the LLM function which will handle streaming but won't edit files
+  const result = await llmFunction(fullPrompt);
+
+  return {
+    content: result.content,
+    generationId: result.generationId,
+    diffData: {}, // No file changes in ask mode
+  };
+};
+
+/**
+ * Performs AI editing operations using streaming with incremental OT operations
+ */
+export const performAIEditing = async ({
+  prompt,
+  shareDBDoc,
+  llmFunction,
+  runCode,
+}) => {
+  // Capture the current state of files before editing
+  const beforeFiles = createFilesSnapshot(
+    shareDBDoc.data.files,
+  );
+
+  const { files } = prepareFilesForPrompt(
+    shareDBDoc.data.files,
+  );
+  const filesContext = formatMarkdownFiles(files);
+
+  // Assemble the final prompt
+  const fullPrompt = assembleFullPrompt({
+    filesContext,
+    prompt,
+    editFormat: 'whole',
+  });
+
+  if (delayStart) {
+    await new Promise((resolve) =>
+      setTimeout(resolve, 1000),
+    );
+  }
+
+  // Call the LLM function which will handle streaming, incremental file updates, and Prettier formatting
+  const result = await llmFunction(fullPrompt);
+
+  runCode();
+
+  // 4. Capture the state of files after editing and formatting, then generate diff
+  const afterFiles = createFilesSnapshot(
+    shareDBDoc.data.files,
+  );
+  const diffData = generateFilesUnifiedDiff(
+    beforeFiles,
+    afterFiles,
+  );
+
+  return {
+    content: result.content,
+    generationId: result.generationId,
+    diffData,
+  };
+};