Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.getbindu.com/llms.txt

Use this file to discover all available pages before exploring further.

Here’s the deal. You’ve got bindu installed and an OPENAI_API_KEY in .env (or exported in your shell). The block below is the whole thing — copy it into agent.py, run it, and 60 seconds later you have a live agent at http://localhost:3773. 
from dotenv import load_dotenv
from bindu.penguin.bindufy import bindufy
from agno.agent import Agent
from agno.models.openai import OpenAIChat
from agno.tools.duckduckgo import DuckDuckGoTools

load_dotenv()  # picks up OPENAI_API_KEY from .env

agent = Agent(
    instructions="You are a research assistant.",
    model=OpenAIChat(id="gpt-4o"),
    tools=[DuckDuckGoTools()],
)

config = {
    "author": "you@example.com",
    "name": "research_agent",
    "description": "Research assistant with web search.",
    "deployment": {"url": "http://localhost:3773", "expose": True},
}

def handler(messages: list[dict[str, str]]):
    return agent.run(input=messages)

bindufy(config, handler)
# To put it on the public internet via FRP:
# bindufy(config, handler, launch=True)

uv run agent.py
You should see a sunflower in your terminal and a DID that looks like did:bindu:you_at_example_com:research_agent:<id>. The @ and . in your email get sanitized to _at_ and _ (source) so the DID stays URL-safe. That’s your agent. It’s online.

​
What just happened

Four things are doing work in those 20 lines. None of them are infrastructure you had to write. agent is your brain. Agno here, but it could be LangChain, CrewAI, a bare OpenAI call, or your own loop. Bindu doesn’t care — it never opens the box. config is the agent’s passport. The author and name get baked into a DID — a cryptographic identity nobody can fake. The deployment.expose key is required (not optional): it declares whether the agent is intended to be publicly reachable (True) or local-only (False). The actual public-internet tunnel is opened by passing launch=True to bindufy() — see the commented line in the snippet. handler is the function the rest of the world will call. It gets a list of messages, returns a response. That’s the entire contract. bindufy(config, handler) wires it all up. Behind it: A2A JSON-RPC routing, DID key generation, Ed25519 signatures on every response, OAuth2 scopes, x402 payment hooks, an A2A-compliant HTTP server on :3773, an optional FRP tunnel when you pass launch=True, and an agent card at /.well-known/agent.json so other agents can discover you. You did not write any of that. You’re welcome.

​
Try it

Open another terminal and send your agent a message: 
curl -X POST http://localhost:3773/ \
  -H 'Content-Type: application/json' \
  -d '{
    "jsonrpc": "2.0",
    "method": "message/send",
    "id": "1",
    "params": {
      "message": {
        "role": "user",
        "kind": "message",
        "parts": [{"kind": "text", "text": "What is x402?"}],
        "messageId": "m1",
        "contextId": "c1",
        "taskId": "t1"
      }
    }
  }'
You get back a taskId. Poll tasks/get with it until state hits completed, and the answer comes out with a DID signature attached. Anyone on the network can verify it really came from your agent.

​
Make it yours

Three things you can change in under a minute:
  1. Swap the prompt. Change instructions= to whatever your agent actually does. Restart.
  2. Swap the model. Replace OpenAIChat(id="gpt-4o") with OpenRouter(id="openai/gpt-oss-120b"), set OPENROUTER_API_KEY in .env. Done — every agent built on Bindu in the wild is on something like this.
  3. Add a tool. Pile more into tools=[...]. The agent will figure out when to call them.

​
What’s next

You have a working agent. The interesting stuff starts when you give it a real job and connect it to others.
  • Config reference — every key in that config dict, what it does, what env var overrides it.
  • Skills — declare what your agent can do so other agents can find it without guessing.
  • Gateway — point a planner at a fleet of agents and let it route work between them.
  • Examples — 30+ agents in this repo doing real things. CSV analysis, PDF Q&A, cybersecurity newsletters, multi-agent swarms.
One sunflower at a time. Welcome to the agent internet.