Thunder — the human engine for AI agents

AI agents browse like robots and get blocked. Thunder gives an agent a human-shaped presence on the web:

Quickstart

pip install thunder-agent
thunder browse https://example.com          # human-shaped browsing
thunder ghost https://nowsecure.nl          # TLS-true fetch
thunder demo                                # full showcase

import thunder, asyncio

async def main():
    body = thunder.Body(profile="ada")            # persistent human
    await body.wake("https://en.wikipedia.org")
    await body.type("#searchInput", "lightning")  # types like a person
    print(await body.read())                     # structured page for your LLM
    await body.rest()

asyncio.run(main())

Pair with ANY model (local or remote)

from thunder.mind import StewMind

mind = StewMind.backend_ollama("stewmind:1.7b")   # fully offline
mind = StewMind(endpoint="http://localhost:1234/v1")  # LM Studio / vLLM
mind = StewMind.backend_huggingface()             # stewmind on HF
mind = StewMind.backend_local()                    # transformers, offline
mind = StewMind(endpoint="https://api.openai.com/v1",
                model="gpt-4o-mini", api_key=KEY)  # any OpenAI-compatible

plan = mind.think(goal="log in and download the invoice",
                  page=await body.read())
# {"action": "type", "target": "#email", "value": "…", "reason": "…"}

Offline pairing: OLLAMA_HOST + OLLAMA_MODEL (or STEWMIND_LOCAL=1) are auto-detected. stewmind-1.7b runs on a laptop CPU/GPU.

MCP — give every agent a human body

Copy-paste configs (Thunder is on PyPI):

{ "mcpServers": { "thunder": {
    "command": "pipx", "args": ["run", "thunder-agent", "thunder-mcp"] } } }

{ "mcpServers": { "thunder": {
    "command": "uvx", "args": ["--from", "thunder-agent", "thunder-mcp"] } } }

Any OpenAI-style config, any software that speaks MCP — Thunder gives it a human body.

Tools: thunder_open, thunder_read, thunder_click, thunder_type, thunder_scroll, thunder_look, thunder_ghost, thunder_captcha, thunder_think, thunder_rest.

CAPTCHAs, honestly

1. Avoid: good skin + human hand + clean IP avoids most challenges.
2. Detect: thunder_captcha reports type + sitekey.
3. Solve: bring your own solver key (CAPSOLVER_API_KEY / TWOCAPTCHA_API_KEY) for your own accounts.
4. Vision (experimental): describe_image() routes challenges to a vision model you choose.

Thunder is not a mass-solving service.

Run it in the cloud (or on your box)

docker build -t thunder .
docker run -p 8000:8000 thunder          # or run natively, fully offline

Responsible use

Thunder is for agents acting on systems and accounts their operator is authorized to use: QA, testing, personal automation, public-data research. Not for fraud, mass account abuse, or deceiving people. Respect site ToS.

MIT © 2026 Emmanuel Ene Rejoice · MUTYINT · Nigeria 🇳🇬