examples

Examples

A working catalog of bindu agents — every one is a single command away from running on your machine, signing its own A2A traffic with a DID, and being reachable on http://localhost:3773. Pick an example, open its folder README, follow the four steps in it.

Each folder has its own README with the specific keys, run command, and curl recipe. This page is just the map.

Prereqs (one-time)

git clone https://github.com/getbindu/Bindu.git && cd Bindu
uv sync --extra agents
export OPENROUTER_API_KEY=<get one at https://openrouter.ai/keys>

OPENROUTER_API_KEY is the single key the whole catalog runs on by default — most examples point their LLM client at OpenRouter regardless of framework, so one key unlocks the Agno, LangChain, AG2, and TypeScript agents alike. A few examples need additional service keys (Mem0, ScrapeGraph, Notte, etc.); their READMEs spell that out. One alternative: beginner/minimax_example.py uses the MiniMax API instead — set MINIMAX_API_KEY (from platform.minimaxi.com) to try that one.

AUTH__ENABLED=true is the default. Every request needs an OAuth token from Hydra plus a DID-signed body — see docs/AUTH.md. To poke around without that ceremony, set AUTH__ENABLED=false in examples/.env and use plain curl.

Python — single agent

The cleanest starting point. Each is one file, one framework, runnable in under a minute.

Folder	What it is	Notable
beginner/	Twelve small agents in one folder — echo, agno, ag2, dspy, paywall variants.	Start here.
summarizer/	Text → 2–3 sentence summary.	Agno + OpenRouter.
weather-research/	Weather report for any city.	DuckDuckGo search.
medical_agent/	Educational medical Q&A with disclaimers baked into the prompt.	DuckDuckGo.
cybersecurity-newsletter/	Topic → headlines + CVEs + brief.	DuckDuckGo.
document-analyzer/	PDF/DOCX → Q&A.	Reads the document from a file part.
pdf_research_agent/	PDF → summary by host path or file part.	pypdf.
speech-to-text/	Audio → transcript.	Gemini 2.0 Flash via OpenRouter.
ai-data-analysis-agent/	CSV → profile + chart.	pandas + matplotlib + seaborn.
news-summarizer/	News headlines + sentiment, local LLM.	Ollama, no API key.
multilingual-collab-agent/	English/Hindi/Bengali, persistent memory.	Mem0.
web-scraping-agent/	URL + prompt → structured JSON.	ScrapeGraph + Mem0.
notte-browser-agent/	Drives a real headless browser.	Notte SDK.

Python — multi-agent

Where you see what bindu was actually built for.

Folder	What it is	Notable
agent_swarm/	Planner → Researcher → Summarizer → Critic → Reflection.	5-stage in-process chain.
ag2_research_team/	researcher / analyst / writer under AutoPattern GroupChat.	AG2 (AutoGen) with LLM-driven speaker selection.
cerina_bindu/	CBT exercise generator: Drafter → SafetyGuardian → ClinicalCritic, supervised.	Real LangGraph StateGraph.
langgraph_blog_writing_agent/	Plan → fan out → reduce, map-reduce style.	LangGraph Send for fan-out.
hermes_agent/	Nous Research's hermes-agent (web/file/code tools) with tiered safety.	Includes call.py — the reference signed-request client.
gateway_test_fleet/	Five agents on local ports + a 13-case test matrix for the Gateway.	What docs/GATEWAY.md uses.

Payments + private surface

Folder	What it is
premium-advisor/	x402 paywall: 0.01 USDC on Base Sepolia per query.
beginner/echo_agent_behind_paywall.py	Echo agent gated by x402.
beginner/agno_paywall_example.py	Agno + paywall.
private_skills_agent/	Public vs /agent/private.json with allowed_dids gating.

TypeScript

Same bindufy(config, handler) shape, written in TypeScript via @bindu/sdk. The SDK launches the Python core in the background; you only see your language.

Folder	What it is
typescript-openai-agent/	OpenAI SDK assistant, pointed at OpenRouter.
typescript-langchain-agent/	LangChain.js research agent.
typescript-langchain-quiz-agent/	LangChain.js quiz generator.

Kotlin

Folder	What it is
kotlin-openai-agent/	Kotlin assistant via sdks/kotlin/, pointed at OpenRouter. Needs JDK 17 + gradle.

Deployment

Folder	What it is
runtime-boxd-agent/	Same vanilla agent code, deploys to a boxd microVM via bindu deploy --runtime=boxd.

Shared bits

Folder	What it is
skills/	Reusable skill.yaml definitions referenced by agents via config["skills"].

Sending a message in 30 seconds

Once an agent is running on :3773 with AUTH__ENABLED=false:

curl -sS http://localhost:3773/ \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"message/send","id":"00000000-0000-0000-0000-000000000004","params":{"message":{"role":"user","parts":[{"kind":"text","text":"hello"}],"kind":"message","messageId":"00000000-0000-0000-0000-000000000001","contextId":"00000000-0000-0000-0000-000000000002","taskId":"00000000-0000-0000-0000-000000000003"},"configuration":{"acceptedOutputModes":["application/json"]}}}'

That returns a Task object. Poll tasks/get with the same taskId for the final artifact. With auth on, sign each body with the agent's DID key — examples/hermes_agent/call.py is the working reference client.

Building your own

Start from the smallest possible shape:

from bindu.penguin.bindufy import bindufy

def handler(messages):
    return f"You said: {messages[-1]['content']}"

bindufy(
    {
        "author": "you@example.com",
        "name": "my_agent",
        "description": "what my agent does",
        "deployment": {"url": "http://localhost:3773", "expose": True},
        "skills": [],
    },
    handler,
)

That's a live agent at http://localhost:3773 with a DID, an agent card at /.well-known/agent.json, A2A, and (by default) auth. Swap in your framework — Agno, LangChain, DSPy, OpenAI SDK directly, your own thing — and you're done.

Beyond this page: docs.getbindu.com for the full reference, docs/ for the deep dives, Discord when you get stuck.