Give any Python agent eyes and hands inside your React app: semantic page observation, cursor-led actions (click, type, scroll, drag), and guided-tour overlays (highlight, callout) — over a WebSocket to the user's live browser tab.
Pairs with the @guidebridge/react
frontend SDK.
from fastapi import FastAPI
from guidebridge import AgentBridge
app = FastAPI()
bridge = AgentBridge()
app.include_router(bridge.router) # mounts ws endpoint at /agent/ws
tools = bridge.as_langchain_tools() # or bridge.as_openai_tools()# LangChain / LangGraph
tools = bridge.as_langchain_tools()
# OpenAI / Anthropic / Google ADK — JSON-schema specs + async dispatcher
toolset = bridge.as_openai_tools()
toolset.specs
await toolset.call("click", '{"target_id": "checkout"}')The agent gets 11 tools — observe_page, point_at, highlight, callout, click,
type_text, select_option, scroll_to, scroll_by, drag, app_action — each
executed in the user's live tab with a visible cursor. Optional extras:
guidebridge[fastapi], guidebridge[langchain].
Full documentation, protocol reference, demo, and security model: github.com/pramodthe/guidebridge.
MIT © Pramod Thebe