ThinkEx-OSS · urjitc · Jul 26, 2026 · Jul 26, 2026 · Jul 26, 2026 · Jul 28, 2026
diff --git a/.dev.vars.example b/.dev.vars.example
@@ -30,14 +30,14 @@ BETTER_AUTH_URL="http://localhost:3000"
 # Autumn usage analytics. When unset, usage tracking is skipped.
 # AUTUMN_SECRET_KEY=""
 
-# Web search and URL/PDF extraction (Firecrawl).
+# Web search and URL extraction (Firecrawl).
 # FIRECRAWL_API_KEY=""
 
 # Agent observability export (The Context Company).
 # TCC_API_KEY=""
 
-# Document extraction (LlamaParse).
-# LLAMA_CLOUD_API_KEY=""
+# Hosted document extraction (FileRouter).
+# FILEROUTER_API_KEY=""
 
 # Direct workspace-upload completion claims. Use a separate random 32+ byte value.
 # WORKSPACE_UPLOAD_TOKEN_SECRET=""
diff --git a/README.md b/README.md
@@ -72,7 +72,7 @@ ThinkEx is a full-stack TypeScript app hosted on Cloudflare. The frontend is Rea
 ### AI, data, and product systems
 
 - **AI SDK**, **Cloudflare Think**, **Cloudflare Sandbox**, **Cloudflare Shell**, and **Cloudflare Codemode**.
-- **Better Auth**, **Drizzle ORM**, **Zod**, **PostHog**, **The Context Company**, **Autumn**, **Firecrawl**, and **LlamaCloud** integrations.
+- **Better Auth**, **Drizzle ORM**, **Zod**, **PostHog**, **The Context Company**, **Autumn**, **Firecrawl**, and **FileRouter** integrations.
 - **Streamdown**, **KaTeX**, **Shiki**, **PapaParse**, **dnd-kit**, **react-resizable-panels**, and **Zustand** for workspace interactions.
 
 See [`package.json`](package.json), [`wrangler.jsonc`](wrangler.jsonc), and [`docs/ENVIRONMENT.md`](docs/ENVIRONMENT.md) for deeper implementation details.

diff --git a/containers/liteparse/package-lock.json b/containers/liteparse/package-lock.json
diff --git a/containers/liteparse/package.json b/containers/liteparse/package.json
diff --git a/containers/liteparse/Dockerfile → ...iners/workspace-file-processor/Dockerfile b/containers/liteparse/Dockerfile → ...iners/workspace-file-processor/Dockerfile
@@ -6,9 +6,6 @@ RUN apt-get update \
 
 WORKDIR /app
 
-COPY package.json package-lock.json ./
-RUN npm ci --omit=dev && npm cache clean --force
-
 COPY server.mjs ./
 
 USER node

diff --git a/containers/liteparse/server.mjs → ...iners/workspace-file-processor/server.mjs b/containers/liteparse/server.mjs → ...iners/workspace-file-processor/server.mjs
@@ -1,6 +1,4 @@
-import { LiteParse } from "@llamaindex/liteparse";
 import { execFile } from "node:child_process";
-import { once } from "node:events";
 import { createWriteStream } from "node:fs";
 import { mkdtemp, readFile, rm } from "node:fs/promises";
 import { createServer } from "node:http";
@@ -11,21 +9,12 @@ import { Transform } from "node:stream";
 import { promisify } from "node:util";
 
 const port = 8080;
-const parser = new LiteParse({
-	extractLinks: true,
-	imageMode: "placeholder",
-	ocrEnabled: false,
-	outputFormat: "markdown",
-	quiet: true,
-});
-const parseTimeoutMs = 90_000;
 const maxInputBytes = 100 * 1024 * 1024;
 const execFileAsync = promisify(execFile);
 
 createServer(async (request, response) => {
 	const startedAt = Date.now();
 	let inputBytes = 0;
-	let pageCount = 0;
 	let status = 500;
 	let errorType = null;
 	let errorMessage = null;
@@ -48,46 +37,26 @@ createServer(async (request, response) => {
 			return response.end(preview.bytes);
 		}
 
-		if (request.method !== "POST" || request.url !== "/parse/pdf") {
-			status = 404;
-			return sendJson(response, status, { error: "Not found." });
-		}
-
-		const bytes = await readPdfRequestBytes(request);
-		inputBytes = bytes.byteLength;
-		const result = await withTimeout(parser.parse(bytes), parseTimeoutMs);
-		pageCount = result.pages.length;
-		status = 200;
-		response.writeHead(status, { "content-type": "application/x-ndjson; charset=utf-8" });
-		for (const page of result.pages) {
-			if (
-				!response.write(
-					`${JSON.stringify({ markdown: page.markdown, pageNumber: page.pageNum })}\n`,
-				)
-			) {
-				await once(response, "drain");
-			}
-		}
-		return response.end();
+		status = 404;
+		return sendJson(response, status, { error: "Not found." });
 	} catch (error) {
 		errorType = error instanceof Error ? error.name : "UnknownError";
 		errorMessage = error instanceof Error ? error.message : String(error);
 		if (error instanceof PdfValidationError) {
 			status = error.status;
 			return sendJson(response, status, { code: error.code, error: error.message });
 		}
-		return sendJson(response, status, { error: "PDF parsing failed." });
+		return sendJson(response, status, { error: "File preview failed." });
 	} finally {
 		console.info(
 			JSON.stringify({
 				duration_ms: Date.now() - startedAt,
 				error_type: errorType,
 				error_message: errorMessage,
-				event: "liteparse_request",
+				event: "workspace_file_preview_request",
 				input_bytes: inputBytes,
 				method: request.method,
 				outcome: status < 400 ? "success" : "error",
-				page_count: pageCount,
 				path: request.url,
 				status,
 			}),
@@ -113,14 +82,6 @@ async function validatePdfFile(filePath) {
 	}
 }
 
-async function readPdfRequestBytes(request) {
-	return withRequestFile(
-		request,
-		"thinkex-parse-",
-		async ({ filePath }) => new Uint8Array(await readFile(filePath)),
-	);
-}
-
 async function generatePreviewRequest(request, kind) {
 	return withRequestFile(request, "thinkex-preview-", async ({ filePath, sizeBytes, tempDir }) => {
 		if (kind === "pdf") {
@@ -202,16 +163,3 @@ function sendJson(response, status, body) {
 	response.writeHead(status, { "content-type": "application/json" });
 	response.end(JSON.stringify(body));
 }
-
-function withTimeout(promise, timeoutMs) {
-	let timeout;
-	const timeoutPromise = new Promise((_, reject) => {
-		timeout = setTimeout(() => {
-			reject(new Error(`LiteParse parsing timed out after ${timeoutMs}ms.`));
-		}, timeoutMs);
-	});
-
-	return Promise.race([promise, timeoutPromise]).finally(() => {
-		clearTimeout(timeout);
-	});
-}
diff --git a/docs/ENVIRONMENT.md b/docs/ENVIRONMENT.md
@@ -38,9 +38,9 @@ The dev server boots and lets you sign in with just the two variables below. Eve
 | `GOOGLE_CLIENT_ID` / `GOOGLE_CLIENT_SECRET` | Google sign-in | Google sign-in fails; use **Continue as guest** (shown in local dev) |
 | `AI_GATEWAY_API_KEY` | AI chat / title generation (Vercel AI Gateway) | AI calls error when invoked |
 | `AUTUMN_SECRET_KEY` | Autumn usage analytics for completed AI chat messages | Usage tracking is skipped |
-| `FIRECRAWL_API_KEY` | Web search + URL/PDF extraction (Firecrawl) | Those calls error when invoked |
+| `FIRECRAWL_API_KEY` | Web search and URL extraction (Firecrawl) | Those calls error when invoked |
 | `TCC_API_KEY` | Agent observability export (The Context Company) | AI chat still works; TCC export is skipped |
-| `LLAMA_CLOUD_API_KEY` | Document extraction (LlamaParse) | Those calls error when invoked |
+| `FILEROUTER_API_KEY` | Hosted document extraction (FileRouter) | File extraction errors when invoked |
 
 ## Signing in without Google
 

diff --git a/docs/concepts/items.mdx b/docs/concepts/items.mdx
@@ -40,4 +40,4 @@ Upload limits are currently 50 files or 100 MB per selection.
 
 ## Extraction and Previews
 
-PDFs and images are routed through extraction and preview pipelines so the AI can read them and the UI can render usable previews. The current stack uses Cloudflare Workflows, R2, Workers AI, LlamaParse, and conversion containers depending on the file type and configured secrets.
+PDFs and images are routed through extraction and preview pipelines so the AI can read them and the UI can render usable previews. The current stack uses FileRouter, Cloudflare Workflows, R2, Workers AI, and conversion containers depending on the file type and configured secrets.
diff --git a/docs/configuration/architecture.mdx b/docs/configuration/architecture.mdx
@@ -52,8 +52,8 @@ The public landing page and product app are currently served by the same Worker.
 | --- | --- |
 | Better Auth | Sessions, Google sign-in, local guest sign-in |
 | AI Gateway / AI SDK | Chat models and title generation |
-| Firecrawl | Public web search and URL/PDF extraction |
-| LlamaCloud | Document extraction |
+| Firecrawl | Public web search and URL extraction |
+| FileRouter | Hosted document extraction with LiteParse and LlamaParse |
 | The Context Company | Agent observability export |
 | Autumn | AI usage analytics |
 | PostHog | Product analytics, feedback, and diagnostics |

diff --git a/docs/configuration/environment.mdx b/docs/configuration/environment.mdx
@@ -19,9 +19,9 @@ ThinkEx reads runtime configuration from Cloudflare bindings, environment variab
 | `GOOGLE_CLIENT_ID` / `GOOGLE_CLIENT_SECRET` | Google sign-in |
 | `AI_GATEWAY_API_KEY` | AI chat and title generation |
 | `AUTUMN_SECRET_KEY` | AI usage analytics |
-| `FIRECRAWL_API_KEY` | Web search and URL/PDF extraction |
+| `FIRECRAWL_API_KEY` | Web search and URL extraction |
 | `TCC_API_KEY` | Agent observability export |
-| `LLAMA_CLOUD_API_KEY` | Document extraction |
+| `FILEROUTER_API_KEY` | Hosted document extraction |
 
 Declared but unset secrets do not stop local startup. The matching feature fails or stays disabled when invoked.
 

diff --git a/package.json b/package.json
@@ -68,6 +68,7 @@
 		"@embedpdf/plugin-tiling": "^2.14.4",
 		"@embedpdf/plugin-viewport": "^2.14.4",
 		"@embedpdf/plugin-zoom": "^2.14.4",
+		"@file_router/sdk": "0.5.0",
 		"@fontsource-variable/geist": "5.3.0",
 		"@modelcontextprotocol/sdk": "^1.29.0",
 		"@posthog/ai": "^8.4.0",
Original file line number	Diff line number	Diff line change
Expand Up		@@ -40,4 +40,4 @@ Upload limits are currently 50 files or 100 MB per selection.

		## Extraction and Previews

		PDFs and images are routed through extraction and preview pipelines so the AI can read them and the UI can render usable previews. The current stack uses Cloudflare Workflows, R2, Workers AI, LlamaParse, and conversion containers depending on the file type and configured secrets.
		PDFs and images are routed through extraction and preview pipelines so the AI can read them and the UI can render usable previews. The current stack uses FileRouter, Cloudflare Workflows, R2, Workers AI, and conversion containers depending on the file type and configured secrets.