Promote Sign in Sign up

Repository

create-llama-worker

Scaffold a CloudFlare worker for LlamaParse Platform services (Parse, Extract, Classify)

LlamaIndex
Repository
Option
Required
-t, --tier <tier>
yes
-v, --version <version>
yes
-s, --schema <json>
yes

Specs

  • OptionRequired
  • -t, --tier <tier>yes
  • -v, --version <version>yes
  • OptionRequired
  • -s, --schema <json>yes
  • -t, --tier <tier>no
  • -v, --version <version>no
  • -p, --system-prompt <prompt>no
  • OptionRequired
  • -r, --rule <type:description>yes
  • OptionDescription
  • -o, --out-dir <dir>Directory to scaffold the worker into (default: . )
  • --api-key <key>LLAMA_CLOUD_API_KEY to write into .env (defaults to $LLAMA_CLOUD_API_KEY )
  • --base-url <url>LLAMA_CLOUD_BASE_URL to write into .env (defaults to $LLAMA_CLOUD_BASE_URL )
  • --project-name <name>Name of the project

README

create-llama-worker

Scaffold a Cloudflare Worker that exposes a single LlamaParse Platform service — Parse, Extract, or Classify — behind a simple HTTP endpoint.

Point it at a running worker with a file upload and get back structured JSON: markdown/text (Parse), structured data (Extract), or a label with reasoning and confidence (Classify).

Usage

Run without arguments for an interactive setup:

npx @llamaindex/create-llama-worker
npm create @llamaindex/llama-worker

Or use one of the subcommands directly for non-interactive / scripted scaffolding.

parse

Scaffold a worker that runs LlamaParse to turn documents into markdown/text.

npx @llamaindex/create-llama-worker parse \
  --tier agentic \
  --version latest \
  --out-dir my-parse-worker
Option Required Description
-t, --tier <tier> yes fast | cost_effective | agentic | agentic_plus
-v, --version <version> yes latest or a YYYY-MM-DD release date

extract

Scaffold a worker that runs LlamaExtract to pull structured data out of documents.

npx @llamaindex/create-llama-worker extract \
  --schema '{"type":"object","properties":{"invoice_number":{"type":"string"}}}' \
  --tier agentic \
  --system-prompt "Extract invoice fields only" \
  --out-dir my-extract-worker
Option Required Description
-s, --schema <json> yes JSON schema describing the data to extract, as a JSON string
-t, --tier <tier> no cost_effective | agentic
-v, --version <version> no latest or a YYYY-MM-DD release date
-p, --system-prompt <prompt> no Custom system prompt to guide extraction

classify

Scaffold a worker that runs LlamaClassify to label documents against custom rules.

npx @llamaindex/create-llama-worker classify \
  --rule "invoice:A billing document requesting payment" \
  --rule "contract:A legal agreement between two parties" \
  --out-dir my-classify-worker
Option Required Description
-r, --rule <type:description> yes A classification rule as <type>:<description>. Repeat for multiple rules.

Common options

Available on every subcommand:

Option Description
-o, --out-dir <dir> Directory to scaffold the worker into (default: .)
--api-key <key> LLAMA_CLOUD_API_KEY to write into .env (defaults to $LLAMA_CLOUD_API_KEY)
--base-url <url> LLAMA_CLOUD_BASE_URL to write into .env (defaults to $LLAMA_CLOUD_BASE_URL)
--project-name <name> Name of the project

What gets generated

Each run scaffolds a self-contained Cloudflare Worker project in --out-dir:

<out-dir>/
├── src/index.ts        # the Worker: uploads the file, calls the LlamaCloud service, returns JSON
├── .env                # LLAMA_CLOUD_API_KEY / LLAMA_CLOUD_BASE_URL
├── wrangler.jsonc       # Wrangler configuration (worker name, compatibility date, nodejs_compat)
├── package.json         # scripts: dev, deploy, test, cf-typegen, lint, prettier
├── .gitignore
├── .prettierrc
└── eslint.config.mts

After scaffolding:

cd <out-dir>
npm install
npm run cf-typegen   # generate Env types from wrangler.jsonc
npm run dev          # run locally with wrangler dev
npm run deploy       # deploy to Cloudflare

The generated Worker accepts a multipart/form-data POST request with a file field, uploads it to LlamaCloud, runs the configured service, and returns JSON, e.g.:

curl -X POST http://localhost:8787 -F "[email protected]"

Development

This repo (the CLI itself, not the generated workers) uses pnpm.

pnpm install
pnpm run build     # bundles src/index.ts -> dist/index.js with esbuild
pnpm run lint
pnpm run prettier:fix

Source layout:

  • src/index.ts – entry point; runs the interactive TUI when called with no args, otherwise dispatches to the CLI
  • src/cli.ts – commander-based non-interactive CLI (parse / extract / classify subcommands)
  • src/tui.ts – @clack/prompts-based interactive flow
  • src/scaffold.ts – writes the rendered files to disk
  • src/render.ts – renders templates/*.tpl (via nunjucks) into the final generated files
  • templates/ – Nunjucks templates for the generated Worker's index.ts, wrangler.jsonc, package.json, etc.

License

MIT