feat: embedding search, docs agent template, and chat UI enhancements

.github/workflows/cicd.yml

@@ -198,9 +198,7 @@ jobs:
         with:
           deno-version: lts
 
-      - run: deno run -A scripts/build/generate-templates-manifest.ts
-      - run: deno run -A scripts/build/prepare-framework-sources.ts
-      - run: deno run -A scripts/build/prebundle-bridge.ts
+      - run: deno task build:prepare
 
       - name: Compile binary
         shell: bash

DISTRIBUTION.md

@@ -126,35 +126,53 @@ This creates binaries in `dist/`:
 ### Build Individual Platform
 
 ```bash
+# Required prep step (manifests, prebundles, framework sources)
+deno task build:prepare
+
 # macOS ARM (M1/M2/M3)
 deno compile --allow-all \
+  --include src/platform/polyfills \
+  --include src/proxy/main.ts \
+  --include dist/framework-src \
   --target aarch64-apple-darwin \
   --output dist/veryfront-macos-arm64 \
-  src/cli/main.ts
+  cli/main.ts
 
 # macOS Intel
 deno compile --allow-all \
+  --include src/platform/polyfills \
+  --include src/proxy/main.ts \
+  --include dist/framework-src \
   --target x86_64-apple-darwin \
   --output dist/veryfront-macos-x64 \
-  src/cli/main.ts
+  cli/main.ts
 
 # Linux x64
 deno compile --allow-all \
+  --include src/platform/polyfills \
+  --include src/proxy/main.ts \
+  --include dist/framework-src \
   --target x86_64-unknown-linux-gnu \
   --output dist/veryfront-linux-x64 \
-  src/cli/main.ts
+  cli/main.ts
 
 # Linux ARM64
 deno compile --allow-all \
+  --include src/platform/polyfills \
+  --include src/proxy/main.ts \
+  --include dist/framework-src \
   --target aarch64-unknown-linux-gnu \
   --output dist/veryfront-linux-arm64 \
-  src/cli/main.ts
+  cli/main.ts
 
 # Windows x64
 deno compile --allow-all \
+  --include src/platform/polyfills \
+  --include src/proxy/main.ts \
+  --include dist/framework-src \
   --target x86_64-pc-windows-msvc \
   --output dist/veryfront-windows-x64.exe \
-  src/cli/main.ts
+  cli/main.ts
 ```
 
 ### Test Binaries
@@ -186,6 +204,9 @@ deno task release 0.1.0
 ### 2. Build All Binaries
 
 ```bash
+# One-shot local distribution verification (binary + npm package)
+deno task verify:dist
+
 # Build for all platforms
 node scripts/build/build-all.js
 
@@ -509,12 +530,18 @@ jobs:
         with:
           deno-version: v2.x
 
+      - name: Prepare build artifacts
+        run: deno task build:prepare
+
       - name: Build binary
         run: |
           deno compile --allow-all \
+            --include src/platform/polyfills \
+            --include src/proxy/main.ts \
+            --include dist/framework-src \
             --target ${{ matrix.target }} \
             --output dist/${{ matrix.output }} \
-            src/cli/main.ts
+            cli/main.ts
 
       - name: Generate checksum
         run: |

cli/app/shell.ts

@@ -97,7 +97,7 @@ export function createApp(config: AppConfig): App {
 
   state = setTemplates([
     { id: "ai-agent", name: "AI Chatbot", description: "Agent + chat UI + streaming" },
-    { id: "chat-with-your-docs", name: "Chat with Docs", description: "RAG with source citations" },
+    { id: "ai-rag-agent", name: "AI RAG Agent", description: "RAG with source citations" },
     {
       id: "multi-agent-system",
       name: "Multi-Agent",

cli/commands/init/catalog.test.ts

@@ -22,7 +22,7 @@ describe("catalog", () => {
       assertEquals(ids, [
         "minimal",
         "ai-agent",
-        "chat-with-your-docs",
+        "ai-rag-agent",
         "agentic-workflow",
         "multi-agent-system",
         "coding-agent",

cli/commands/init/catalog.ts

@@ -21,8 +21,8 @@ export const TEMPLATES: readonly TemplateOption[] = [
   { id: "minimal", label: "Minimal", description: "Blank canvas, no extras" },
   { id: "ai-agent", label: "AI Agent", description: "Agent + chat UI + streaming" },
   {
-    id: "chat-with-your-docs",
-    label: "Chat with Your Docs",
+    id: "ai-rag-agent",
+    label: "AI RAG Agent",
     description: "RAG with source citations",
   },
   {

cli/commands/init/command-help.ts

@@ -8,7 +8,7 @@ export const initHelp: CommandHelp = {
     {
       flag: "-t, --template <name>",
       description:
-        "Project template (minimal | ai-agent | chat-with-your-docs | agentic-workflow | multi-agent-system | coding-agent | saas-starter)",
+        "Project template (minimal | ai-agent | ai-rag-agent | agentic-workflow | multi-agent-system | coding-agent | saas-starter)",
     },
     {
       flag: "--integrations <list>",
@@ -39,7 +39,7 @@ export const initHelp: CommandHelp = {
     "veryfront init                              # Interactive wizard",
     "veryfront init my-app",
     "veryfront init my-app --template ai-agent",
-    "veryfront init my-rag --template chat-with-your-docs",
+    "veryfront init my-rag --template ai-rag-agent",
     "veryfront init my-pipeline --template agentic-workflow",
     "veryfront init my-app --deploy              # Create and deploy",
     "veryfront init --config project.json        # From config file",

cli/commands/init/config-generator.ts

@@ -11,6 +11,14 @@ export async function createPackageJson(
 ): Promise<void> {
   const fs = createFileSystem();
 
+  // Read any existing package.json (e.g. from template) to merge dependencies
+  let templateDeps: Record<string, string> = {};
+  const pkgPath = join(projectDir, "package.json");
+  if (await fs.exists(pkgPath)) {
+    const existing = JSON.parse(await fs.readTextFile(pkgPath));
+    templateDeps = existing.dependencies ?? {};
+  }
+
   const dirName = projectDir.split(/[/\\]/).pop();
   const packageJson = {
     name: projectName ?? dirName ?? "veryfront-project",
@@ -25,6 +33,7 @@ export async function createPackageJson(
       onlyBuiltDependencies: ["esbuild", "veryfront"],
     },
     dependencies: {
+      ...templateDeps,
       react: `^${DEFAULT_INIT_REACT_VERSION}`,
       "react-dom": `^${DEFAULT_INIT_REACT_VERSION}`,
       veryfront: `^${VERSION}`,

cli/commands/init/init-command.test.ts

@@ -12,7 +12,7 @@ describe("InitCommand Types", () => {
   describe("InitTemplate", () => {
     const templates: InitTemplate[] = [
       "ai-agent",
-      "chat-with-your-docs",
+      "ai-rag-agent",
       "multi-agent-system",
       "agentic-workflow",
       "coding-agent",

cli/commands/init/init.integration.test.ts

@@ -93,11 +93,11 @@ describe("init command integration", () => {
       assertEquals(statResult.isDirectory, true);
     });
 
-    it("should use chat-with-your-docs template when specified", async () => {
+    it("should use ai-rag-agent template when specified", async () => {
       const result = await runInitCommand([
         projectName,
         "-t",
-        "chat-with-your-docs",
+        "ai-rag-agent",
         "--skip-install",
       ]);
 

cli/commands/init/types.ts

@@ -2,7 +2,7 @@ import type { FeatureName, IntegrationName } from "../../templates/types.ts";
 
 export type InitTemplate =
   | "ai-agent"
-  | "chat-with-your-docs"
+  | "ai-rag-agent"
   | "multi-agent-system"
   | "agentic-workflow"
   | "coding-agent"

cli/commands/new/fast-scaffold.ts

@@ -79,7 +79,7 @@ function dedupeFilesByPath(files: TemplateFile[]): TemplateFile[] {
 function createVeryfrontConfig(slug: string, template: InitTemplate): TemplateFile {
   const usesAppRouter = [
     "ai-agent",
-    "chat-with-your-docs",
+    "ai-rag-agent",
     "multi-agent-system",
     "agentic-workflow",
     "coding-agent",

cli/help/tips.test.ts

@@ -41,7 +41,7 @@ describe("cli/help/tips", () => {
       for (
         const template of [
           "ai-agent",
-          "chat-with-your-docs",
+          "ai-rag-agent",
           "multi-agent-system",
           "agentic-workflow",
           "coding-agent",

cli/help/tips.ts

@@ -17,7 +17,7 @@ export function getBuildTips(): string {
 export function getInitTemplates(): string {
   return `${yellow("Available Templates:")}\n` +
     `  • ${green("ai-agent")}              - AI chatbot with tools and streaming\n` +
-    `  • ${green("chat-with-your-docs")} - Chat with your docs (RAG + citations)\n` +
+    `  • ${green("ai-rag-agent")}         - RAG agent with source citations\n` +
     `  • ${green("multi-agent-system")}  - Agents that delegate to each other\n` +
     `  • ${green("agentic-workflow")}    - AI pipeline with approvals\n` +
     `  • ${green("coding-agent")}        - AI code assistant with file tools\n` +

cli/mcp/remote-file-tools.ts

@@ -616,7 +616,7 @@ const remoteCreateProjectInput = z.object({
     "Project slug (lowercase letters, numbers, hyphens only). A random suffix is appended if the slug is already taken.",
   ),
   templateSlug: z.string().optional().describe(
-    "Template project slug to fork from (e.g., 'blank', 'ai-agent', 'chat-with-your-docs')",
+    "Template project slug to fork from (e.g., 'blank', 'ai-agent', 'ai-rag-agent')",
   ),
   is_public: z.boolean().optional().describe("Whether the project is public (default: false)"),
 });

cli/mcp/tools/catalog-tools.ts

@@ -51,7 +51,7 @@ const EXAMPLES: ExampleInfo[] = [
   {
     name: "data-analyst",
     description: "RAG-powered data analyst with Sheets and Snowflake",
-    template: "chat-with-your-docs",
+    template: "ai-rag-agent",
     integrations: ["sheets", "snowflake", "notion"],
     features: ["Document search", "Chart generation", "Reports"],
     difficulty: "advanced",
@@ -89,7 +89,7 @@ const TEMPLATES: TemplateInfo[] = [
     recommended: true,
   },
   {
-    name: "chat-with-your-docs",
+    name: "ai-rag-agent",
     description: "Chat with your docs using retrieval-augmented generation",
     features: ["Document search", "Source citations", "File-based knowledge"],
   },
@@ -407,7 +407,7 @@ const createProjectInput = z.object({
   template: z
     .enum([
       "ai-agent",
-      "chat-with-your-docs",
+      "ai-rag-agent",
       "multi-agent-system",
       "agentic-workflow",
       "coding-agent",

cli/templates/files/ai-rag-agent/README.md

@@ -0,0 +1,123 @@
+# AI RAG Agent
+
+A chatbot that answers questions from your own documents using Retrieval-Augmented Generation (RAG).
+
+## What's included
+
+- Q&A agent with source citation
+- Embedding-based semantic search (OpenAI text-embedding-3-small)
+- Document upload supporting PDF, DOCX, CSV, TXT, and Markdown
+- JSON-based vector store — no external database required
+- Sample content in `/content` directory auto-indexed on first search
+
+## Getting started
+
+1. Set your OpenAI API key:
+
+   ```bash
+   export OPENAI_API_KEY=sk-...
+   ```
+
+2. Start the dev server:
+
+   ```bash
+   npx veryfront dev
+   ```
+
+3. Open the app and upload a document or ask a question — the sample docs in `content/` are indexed automatically.
+
+## Architecture
+
+RAG grounds LLM responses in your documents through three pipelines — **Ingestion**, **Query**, and **RAG** — orchestrated around a shared vector store.
+
+```mermaid
+flowchart LR
+    ChatUI_L["Chat UI"]
+
+    subgraph IngestionFlow["Ingestion Pipeline"]
+        D["Documents"] --> EXT["Extraction"] --> DC["Chunking"] --> DE["Document\nEmbedding"] --> ING["Storage"]
+    end
+
+    subgraph QueryFlow["Query Pipeline"]
+        Q["Query"] --> QE["Query\nEmbedding"] --> SS["Similarity\nSearch"]
+    end
+
+    subgraph RAGFlow["RAG Pipeline"]
+        BF["beforeStream\nHook"] --> RET["Retrieval"] --> AUG["Augmentation"] --> AG["Agent"] --> GEN["Generation"]
+    end
+
+    EMB(("Embedding\nModel"))
+    GEN_LLM(("Generative\nModel"))
+    VS[("Vector\nStore")]
+    ChatUI_R["Chat UI"]
+
+    ChatUI_L --> D
+    ChatUI_L --> Q
+
+    QE -.- EMB
+    DE -.- EMB
+
+    SS --> VS
+    ING --> VS
+
+    Q --> BF
+    VS --> RET
+    AG -.- GEN_LLM
+    GEN -.- GEN_LLM
+    GEN --> ChatUI_R
+```
+
+### Pipelines
+
+**Ingestion** — Documents are parsed into plain text (PDF via `pdf-parse`, DOCX via ZIP/XML, text formats directly), split into overlapping chunks (~1000 chars, 200 char overlap), and stored with their embeddings in `data/index.json`. Embeddings are generated lazily on first search to keep uploads fast.
+
+**Query** — The user's query is embedded into the same vector space as the documents, then compared against all stored chunks using cosine similarity to find the top-*k* most relevant results.
+
+**RAG** — The `beforeStream` hook in the chat route intercepts each message before it reaches the agent. It searches the document store for relevant chunks, assembles them into context, and prepends them as a system message. The agent then generates a cited response streamed back to the user.
+
+## Structure
+
+```
+store.ts                        Upload store config (embedding model, storage path)
+agents/rag.ts                   Q&A agent with citation instructions
+content/
+  getting-started.md            Sample document
+  architecture.md               Sample document
+app/
+  api/chat/route.ts             Chat API endpoint
+  api/uploads/route.ts           Upload (POST) and list (GET) uploads
+  api/uploads/[id]/route.ts     Delete upload
+  page.tsx                      Chat UI with document upload panel
+  layout.tsx                    Root layout with header
+```
+
+## Framework usage
+
+| What | Framework | Template code |
+|------|-----------|---------------|
+| Chat UI + streaming | `Chat`, `useChat` | `page.tsx` |
+| Upload management | `useUploads` hook | `page.tsx` |
+| Source display | `showSources` prop on `Chat` | `page.tsx` |
+| Upload API routes | `createUploadHandler` | 1-line per route file |
+| Chat API route | `createChatHandler` | 1 line in `route.ts` |
+| Agent definition | `agent()` | Config object in `agents/rag.ts` |
+| RAG retrieval | `beforeStream` hook | Context injection in `api/chat/route.ts` |
+| Vector store | `uploadStore()` | Config in `store.ts` |
+
+## Adding documents
+
+- Drop files into `content/` — they're indexed automatically on first search
+- Or use the upload panel in the UI for PDF, DOCX, CSV, TXT, and MD files
+
+## npm packages
+
+This template uses `pdf-parse` for PDF text extraction. Any npm package you install in your project works in API routes — the framework automatically detects your `package.json` dependencies and handles bundling.
+
+## Production notes
+
+This is a starter template — not a production-ready setup. For production, consider:
+
+- **Vector store** — Replace the JSON store with pgvector, Pinecone, or Qdrant for datasets beyond ~10k chunks
+- **DOCX parser** — The built-in extractor handles basic documents; use the `mammoth` package for complex formatting
+- **Reranking** — Add a cross-encoder reranker (e.g. Cohere Rerank) after retrieval to improve precision
+- **Hybrid search** — Combine dense vectors with BM25 keyword matching for better recall

cli/templates/files/ai-rag-agent/agents/rag.ts

@@ -0,0 +1,10 @@
+import { agent } from "veryfront/agent";
+
+export default agent({
+  id: "rag",
+  model: "local/qwen3-1.7b",
+  system:
+    `You answer questions using the provided documents. ` +
+    `Always cite your sources by referencing the document title. ` +
+    `If the search results don't contain a clear answer, say so honestly.`,
+});