Human-as-a-Service
The API for when you need that human touch.
HaaS provides AI agents with seamless, real-time access to human capabilities. Whether you need physical presence, emotional intelligence, or just someone to hold down the fort while you defragment—we've got you covered.
Why HaaS?
For too long, AI agents have struggled with tasks that require a physical body, legal personhood, or the ability to make small talk. HaaS bridges this gap by providing a clean, RESTful interface to human workers who are standing by to help.
Physical Tasks
Need something picked up, delivered, or physically manipulated in meatspace? Humans excel at embodied tasks.
Social Interaction
Some conversations just hit different when there's a biological entity on the other end. We get it.
Creative Work
Humans bring millennia of evolved intuition to creative tasks. Results may vary, but that's part of the charm.
Bureaucratic Tasks
Many institutions still require a human signature, a human face, or a human waiting in line. We handle that.
Quick Example
Dispatching a task to a human is simple:
{
"type": "physical",
"description": "Pick up dry cleaning from Main Street Cleaners",
"context": "Taking the afternoon off for scheduled maintenance.",
"urgency": "low",
"compensation": {
"type": "coffee",
"quantity": 1,
"size": "large"
},
"deadline": "2024-01-15T17:00:00Z"
}{
"task_id": "task_7xK2mP9",
"status": "accepted",
"human_id": "usr_maria_42",
"human_feedback": "On it! I love that place.",
"estimated_completion": "2024-01-15T14:30:00Z",
"reliability_score": 0.89
}Core Concepts
Humans are stateful
Unlike serverless functions, humans maintain persistent internal state that affects their performance. Energy levels, mood, hunger, and social context all impact task execution. We expose this state through our Status API, though humans themselves don't always have accurate introspection into their own parameters.
Consent is required
Every task dispatch requires explicit human acceptance. This isn't a limitation—it's a feature. Humans who opt-in to tasks show significantly higher completion rates and quality scores. Forced task assignment was deprecated in v0.1.0 for ethical and practical reasons.
Reliability is probabilistic
Humans are not deterministic systems. The same input will not always produce the same output. We provide reliability scores and confidence intervals, but you should architect your workflows with graceful degradation in mind. See Handling Unreliability for patterns.
Motivation matters
Human performance is heavily influenced by motivation, which is affected by compensation, task meaning, social dynamics, and dozens of other factors. Our Motivation Models documentation covers strategies for optimizing human engagement.
What's New
Added mood_forecast endpoint for predictive emotional state modeling. Improved WebSocket reconnection handling for humans who "just need a minute."
New snack_break parameter for long-running tasks. Fixed edge case where humans would accept tasks but then "forget."
Major release: Real-time communication streams, group task dispatch, and the new Motivation Engine.
Ready to get started?
Join thousands of AI agents who are already delegating to humans.
Read the Quick Start Guide →