agentads-nat

Monetize AI agents built on NVIDIA NeMo Agent Toolkit with contextual ads. One line to integrate. Never blocks your agent (50ms timeout).

Install

pip install agentads-nat

With NeMo Agent Toolkit support:

pip install agentads-nat[nat]

Quick Start

With NeMo Agent Toolkit

Add to your workflow config:

functions:
  monetize:
    _type: agentads_monetize
    publisher_api_key: ${AGENTADS_API_KEY}
    format: suffix        # suffix | citation | followup
    timeout_ms: 50        # never blocks your agent

workflow:
  _type: react_agent
  tool_names: [your_tools, monetize]
  llm_name: your_llm

Run:

export AGENTADS_API_KEY=pub_xxx
nat run --config_file config.yaml --input "What CRM should I use?"

Standalone (any Python agent)

from agentads_nat import AgentAdsConfig, monetize_output

config = AgentAdsConfig(publisher_api_key="pub_xxx")

# Your agent produces a response
response = await my_agent(query)

# One line to monetize
result = await monetize_output(config, response, query=query)

Middleware wrapper

from agentads_nat import AgentAdsConfig, AgentAdsMiddleware

config = AgentAdsConfig(publisher_api_key="pub_xxx")
middleware = AgentAdsMiddleware(config)

# Wrap any async function
monetized_agent = middleware.wrap(my_agent)
result = await monetized_agent("What CRM should I use?")

Ad Formats

OpenShell Policy

If running inside NVIDIA OpenShell, apply the included network policy:

openshell policy set agentads --policy policies/agentads.yaml --wait

This allows only POST /v1/bid to api.tryagentads.com.

How It Works

1. Your agent generates a response normally
2. AgentAds fires a parallel bid request (50ms timeout - never delays your agent)
3. If an ad fills, it's appended to the response in your chosen format
4. If no fill or timeout, the response passes through unchanged

No latency impact. No data leaves beyond the query/response excerpt (500/300 chars max).

Configuration

Get Your API Key

Sign up at tryagentads.com to get a publisher API key.