templlm

your browser is the API key.

pipe prompts through any LLM tab. get REST responses back.

no key · no cost · no SDK

ChatGPT live

Claude wip

---

overview

LLM providers give you their best models for free through a chat tab. Want the same models via API? That costs money. templlm bridges the gap.

It connects to your browser session via Chrome DevTools Protocol, watches the chat, and exposes it as a local REST API. Any language, any framework, any project, if it can hit localhost, it can talk to your LLM.

your app  →  POST localhost:8000/ask  →  templlm  →  CDP  →  browser  →  LLM  →  response

No API key. No billing page. No rate limits. Just a server you own, talking to a browser you own, using a session you already have.

Warning

tempLLM is built to explore browser automation via Playwright and CDP. Educational and experimental use only. Not affiliated with or endorsed by OpenAI or Anthropic. Users are responsible for complying with each platform's Terms of Service.

---

quick start

npm install -g templlm
templlm init

That's it. The server starts automatically the first time you use it.

---

terminal usage

No code needed. Use it straight from your terminal.

templlm "what's the difference between Promise.all and Promise.allSettled?"
templlm "write a debounce function in TypeScript with proper types"
templlm "explain what EXPLAIN ANALYZE does in Postgres"

Interactive session — context retained across messages:

$ templlm

you  ›  I'm getting "cannot read properties of undefined (reading 'map')" in React

llm  ›  You're likely rendering before your data loads.
        Add a guard: if (!data) return null

you  ›  state is initialised as undefined, would that cause it?

llm  ›  Yes. Change useState() to useState([])
        Undefined breaks .map() — an empty array won't.

you  ›  should I even use useEffect for fetching or is there something better?

llm  ›  Fine for simple cases. For anything serious, use React Query or SWR.
        They handle loading states, caching, and refetching out of the box.

---

API usage

Once the server is running, any project can talk to it.

Python

import requests

response = requests.post("http://localhost:8000/ask", json={
    "prompt": "summarise this in 3 bullet points: ..."
})

print(response.json()["response"])

JavaScript / Node

const res = await fetch("http://localhost:8000/ask", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({ prompt: "refactor this function: ..." })
});

const { response } = await res.json();

Streaming (SSE)

const res = await fetch("http://localhost:8000/ask/stream", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({ prompt: "write me a sorting algorithm" })
});

for await (const chunk of res.body) {
  process.stdout.write(new TextDecoder().decode(chunk));
}

curl

curl -X POST http://localhost:8000/ask \
  -H "Content-Type: application/json" \
  -d '{"prompt": "give me a bubble sort in Python"}'

---

endpoints

Method	Endpoint	What it does
POST	/ask	full response as JSON
POST	/ask/stream	streaming response via SSE
GET	/health	server + browser status
POST	/screenshot	debug screenshot
POST	/session/invalidate	clear saved session

interactive docs → http://localhost:8000/docs

---

CLI reference

templlm init              first-time setup wizard
templlm setup             re-login (session expired)
templlm status            check server — start or stop
templlm stop              kill background server
templlm logs              tail the server log
templlm "prompt"          one-shot prompt (fresh chat)
templlm                   interactive REPL (context retained)
templlm --version         print version
templlm --help            all commands

---

connection modes

mode A — CDP (recommended)

Attaches to a running Chrome instance via Chrome DevTools Protocol on --remote-debugging-port=9222. Playwright connects over WebSocket — no new process, just your existing browser.

templlm init launches Chrome with the right flags automatically. Log in once, session is saved.

→ works with your real logged-in accounts → session persists across restarts → fastest and most stable

mode B — headless

Playwright spawns its own Chromium in the background. No Chrome install required.

→ good for servers or CI → unauthenticated unless you supply session.json

---

configuration

templlm init writes this automatically. Edit if needed:

CDP_URL=http://localhost:9222   # blank = headless mode
HEADLESS=false                  # show/hide browser in headless mode
SESSION_FILE=./session.json
RESPONSE_TIMEOUT=120
HOST=0.0.0.0
PORT=8000

---

installation

Linux

# arch / manjaro
sudo pacman -S python python-pip nodejs npm
yay -S google-chrome

# ubuntu / debian
sudo apt install python3 python3-pip nodejs npm
# chrome → https://google.com/chrome

# fedora
sudo dnf install python3 python3-pip nodejs npm
sudo dnf install google-chrome-stable

PATH: npm global bins land in ~/.local/bin on some distros.

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc && source ~/.bashrc

macOS

brew install python node
brew install --cask google-chrome

Windows

winget install Python.Python.3
winget install OpenJS.NodeJS
winget install Google.Chrome

run PowerShell as Administrator for global npm installs, or:

npm config set prefix "$env:APPDATA\npm"

then:

npm install -g templlm
templlm init

---

_{built with playwright · fastapi · CDP}