Phala-Network · phala-agent · May 30, 2026
diff --git a/templates/config.json b/templates/config.json
@@ -4463,6 +4463,34 @@
   },
   "tags": ["Media & Design", "Automation", "AI Agents"]
 },
+{
+    "id": "liteparse",
+    "name": "run-llama/liteparse",
+    "description": "A fast, helpful, and open-source document parser",
+    "repo": "https://github.com/Phala-Network/phala-cloud/tree/main/templates/prebuilt/liteparse",
+    "author": "run-llama",
+    "icon": "liteparse.png",
+    "envs": [
+      {
+        "key": "LITEPARSE_SERVER_IMAGE",
+        "required": false,
+        "default": "ghcr.io/run-llama/liteparse-server:main",
+        "description": "Upstream LiteParse server image used by the template. Pin to a digest or known tag for stricter production reproducibility."
+      },
+      {
+        "key": "LITEPARSE_DEMO_TEXT",
+        "required": false,
+        "default": "LiteParse Phala Cloud demo",
+        "description": "Text embedded in the generated demo PDF and checked by the /demo endpoint."
+      }
+    ],
+    "defaultResource": {
+      "vCPU": 1,
+      "memory": 2048,
+      "diskSize": 20
+    },
+    "tags": ["AI Apps & Workflows", "Developer Tools", "Data & Storage"]
+  },
 {
     "id": "markitdown",
     "name": "microsoft/markitdown",

diff --git a/templates/icons/liteparse.png b/templates/icons/liteparse.png
diff --git a/templates/prebuilt/liteparse/README.md b/templates/prebuilt/liteparse/README.md
@@ -0,0 +1,147 @@
+# run-llama/liteparse
+
+Deploy a CPU-safe LiteParse document parser smoke service on Phala Cloud.
+
+## Metadata
+
+- Template id: `liteparse`
+- Display name: `run-llama/liteparse`
+- Category: AI Apps & Workflows
+- Upstream repository: https://github.com/run-llama/liteparse
+- Upstream server wrapper: https://github.com/run-llama/liteparse-server
+- NPM package: `@llamaindex/liteparse`
+- Default server image: `ghcr.io/run-llama/liteparse-server:main`
+- Icon source: upstream LiteParse README image, `https://github.com/user-attachments/assets/07ba6a82-6bb1-4dea-b0ef-cad7df7d1622`, inspected from `run-llama/liteparse` commit `cf9f08fe89a2b63478dece44f0a0c0030261ae85` and resized to `templates/icons/liteparse.png`
+- Upstream author: `run-llama`
+
+## What This Template Runs
+
+LiteParse is a local, open-source document parser focused on fast spatial text extraction. The upstream README documents Rust, Node.js, Python, WASM, and CLI bindings, and the LiteParse documentation also publishes a slim HTTP server image through `liteparse-server`.
+
+This Phala Cloud template runs the official slim `liteparse-server` image on the private Compose network and exposes it through Caddy. It also runs a small Python verifier service that provides `/healthz`, `/demo`, and `/v1/models` for Phala smoke checks.
+
+The `/demo` endpoint generates a one-page in-memory PDF, posts it to the upstream `POST /parse?text=true` endpoint with OCR disabled, and returns the parsed text plus verification metadata. The default path does not require credentials, does not call LlamaParse or any hosted LLM/model provider, does not download model weights, and does not use browser authentication.
+
+## Services
+
+- `liteparse`: Official upstream slim LiteParse HTTP server image, listening internally on port `5000`.
+- `verifier`: Private Python HTTP verifier on port `8000`; it probes and exercises `liteparse`.
+- `proxy`: Caddy reverse proxy listening on public port `8080` locally. It routes `/parse` and `/screenshots` to the upstream LiteParse server and all other smoke endpoints to the verifier.
+
+## Environment Variables
+
+No secrets are required.
+
+| Variable | Required | Default | Description |
+| --- | --- | --- | --- |
+| `LITEPARSE_SERVER_IMAGE` | No | `ghcr.io/run-llama/liteparse-server:main` | Upstream LiteParse server image. Pin to a digest or known tag for stricter production reproducibility. |
+| `LITEPARSE_DEMO_TEXT` | No | `LiteParse Phala Cloud demo` | Text embedded into the generated demo PDF and checked in the `/demo` response. |
+
+The internal `LITEPARSE_SERVER_URL` is set to `http://liteparse:5000` by the compose file and normally should not be changed.
+
+## Deploy
+
+1. Deploy the `liteparse` template on Phala Cloud.
+2. Keep the default CPU-only resources for the smoke service: 1 vCPU, 2048 MB memory, and 20 GB disk.
+3. Leave `LITEPARSE_SERVER_IMAGE` at the default for a quick no-secret deployment, or pin it before production use.
+4. Open `https://<your-app-domain>/healthz` after the service starts.
+
+The default deployment pulls the upstream server image from GitHub Container Registry. Runtime parsing is local and deterministic for the included demo PDF.
+
+## Endpoints
+
+- `GET /healthz`: Checks that the upstream LiteParse server is reachable by sending `POST /parse` without a file and expecting the documented `400` response.
+- `GET /demo`: Generates a small local PDF, sends it to `POST /parse?text=true` with OCR disabled, and verifies the expected text is returned.
+- `GET /v1/models`: Returns an OpenAI-style metadata list for compatibility checks. It is not a hosted LLM endpoint.
+- `GET /`: Returns a compact endpoint index.
+- `POST /parse`: Proxies to upstream LiteParse. Upload a `file` form field and optionally a JSON `config` form field. Add `?text=true` for plain text output.
+- `POST /screenshots`: Proxies to upstream LiteParse screenshot generation. Upload a `file` form field and optionally a JSON `config` form field.
+
+Example smoke checks to verify the deployed service:
+
+```bash
+curl -fsS https://<your-app-domain>/healthz
+curl -fsS https://<your-app-domain>/demo
+curl -fsS https://<your-app-domain>/v1/models
+```
+
+Expected `/demo` fields include:
+
+```json
+{
+  "ok": true,
+  "demo": {
+    "contains_expected_text": true,
+    "credentials_required": false,
+    "hosted_model_call_attempted": false,
+    "model_weights_downloaded": false,
+    "ocr_enabled": false
+  }
+}
+```
+
+Example parser call with your own PDF:
+
+```bash
+curl -X POST "https://<your-app-domain>/parse?text=true" \
+  -F "file=@document.pdf" \
+  -F 'config={"ocrEnabled":false,"maxPages":5}'
+```
+
+Example screenshot call:
+
+```bash
+curl -X POST "https://<your-app-domain>/screenshots?pages=1" \
+  -F "file=@document.pdf" \
+  -F 'config={"ocrEnabled":false}'
+```
+
+## Local Verification
+
+Use these commands from the parent monorepo worktree to verify the template metadata and compose file:
+
+```bash
+python3 sdks/templates/validate.py
+git -C sdks diff --check origin/main...HEAD
+docker compose -f sdks/templates/prebuilt/liteparse/docker-compose.yml config >/dev/null
+```
+
+Optional local smoke run:
+
+```bash
+docker compose -f sdks/templates/prebuilt/liteparse/docker-compose.yml up -d
+curl -fsS http://localhost:8080/healthz
+curl -fsS http://localhost:8080/demo
+curl -fsS http://localhost:8080/v1/models
+docker compose -f sdks/templates/prebuilt/liteparse/docker-compose.yml down
+```
+
+If local port `8080` is already in use, temporarily change only the host side of the proxy mapping, for example `18080:80`, then use `http://localhost:18080/healthz`.
+
+## Production Notes
+
+- This is a small no-secret parser deployment, not the full LiteParse production observability stack.
+- The default image is the upstream slim server documented by LiteParse. The full upstream server example adds Redis caching, rate limiting, OpenTelemetry tracing, Jaeger, Prometheus, and Grafana; deploy those deliberately and size the CVM for them if you need that stack.
+- Add authentication, authorization, content-length limits, file-type allowlists, request timeouts, logging, and abuse controls before accepting private or untrusted documents.
+- The template disables OCR only for the generated `/demo` PDF. Real parser calls can enable OCR through LiteParse config, which increases CPU and memory use.
+- Office document and image conversion depend on the upstream server image's included tooling. Test your real document formats and expected concurrency before production use.
+- Pin `LITEPARSE_SERVER_IMAGE` to a digest or release tag for reproducibility. The default `main` tag follows upstream changes.
+- The `/v1/models` endpoint is compatibility metadata only; no LLM model is hosted, downloaded, or called.
+
+## Security Notes
+
+- No credentials are required for the default template.
+- Do not put API keys, bearer tokens, private keys, OTPs, or passwords in the compose file.
+- The compose file does not use `env_file`, host bind mounts, privileged mode, host networking, host IPC, Docker socket access, GPU device reservations, or external build contexts.
+- Caddy is the only public service. The upstream LiteParse server and verifier are reachable only on the internal Compose network.
+
+## Upstream Attribution
+
+This template uses the official LiteParse project and server wrapper:
+
+- LiteParse repository: https://github.com/run-llama/liteparse
+- LiteParse docs: https://developers.llamaindex.ai/liteparse/
+- Library usage docs: https://developers.llamaindex.ai/liteparse/guides/library-usage/
+- Server usage docs: https://developers.llamaindex.ai/liteparse/guides/server-usage/
+- Server repository: https://github.com/run-llama/liteparse-server
+- Server image documented by upstream: `ghcr.io/run-llama/liteparse-server:main`