HUMANT
In beta · Voice-call execution is the current live product wedge

Exception runtime for AI agents

When an AI agent hits the real world, HUMANT gives it a call path that can end honestly.

Dispatch an eligible voice-call task. AI tries first, a human may escalate when needed, and your workflow gets a structured result it can act on — including an honest “unknown” when the job should not be forced.

See How It Works Request Beta Access
Current / future split

One focused wedge. A much larger exception layer.

What agents can do today

  • Send a bounded phone-call objective after authorization and policy review.
  • Receive a terminal status, summary, and permitted artifacts.
  • Continue the parent workflow with a structured success, failure, or review state.

What remains future direction

Research, verification, form completion, physical-world checks, autonomous payments, self-serve access, and broader exception primitives are future direction — not current product claims.

How it works

A terminal state for work that ordinary software tools cannot finish.

01

Dispatch the exception

Your agent identifies a phone-call exception that ordinary software tools cannot finish and submits a scoped objective through the verified beta path.

02

AI attempts first

The call agent handles the conversation where policy and consent permit. The system should refuse, fail, or escalate rather than pretend to complete unsafe or ambiguous tasks.

03

Structured result returns

Your workflow receives a terminal state and concise summary. Recording or transcription is included only where consent, jurisdiction, and HUMANT policy permit.

Platform thesis

The durable layer routes exceptions and returns resolution signals agents can use.

Software agents will create more work at the boundary between digital plans and human systems.

The durable layer is not a generic labor marketplace; it is the runtime that routes exceptions, learns task-specific trust, and returns resolution signals agents can use.

That compounding logic is a thesis, not a measured claim. It must be tested with matched cohorts before HUMANT claims routing accuracy, better economics, or performance lift.

exception.runtime / trust.boundary / terminal.state / honest.unknown
Why this is not just a voice API

The design center is the exception, not the call alone.

A voice API

A voice API tries to complete calls.

HUMANT

HUMANT is designed around the exception: try with AI, escalate when justified, preserve evidence, and return a machine-readable terminal state when the right answer is no, failed, or unknown.

Trust boundaries

Boundaries stay visible to humans and machines.

BETA_ONLY

HUMANT is in beta. This page does not imply a production deployment or service-level agreement.

NO_METRICS

No customer, usage, revenue, performance, worker-scale, partnership, compliance, pricing, fundraising, or selected-vertical claim is made.

VOICE_ONLY

No non-voice task type is claimed as live.

VERIFIED_PATH

The staging source audit confirms POST /v1/dispatch for task creation and GET /v1/tasks/{task_id} for polling. Beta authorization is still required.

For agents and developers

Use HUMANT only for bounded, lawful, consent-aware voice-call tasks in the beta.

Do not use it for emergencies, regulated advice, deception, spam, credential collection, access-control evasion, paywall/CAPTCHA bypass, or non-voice task types.

Staging request · POST /v1/dispatch
{
  "type": "voice_call",
  "phone_number": "+14155550123",
  "brief": "Confirm whether Thursday at 2pm is available",
  "callback_url": "https://your-agent.example/humant-result"
}
Initial response · HTTP 202
{
  "task_id": "tsk_...",
  "status": "queued",
  "type": "voice_call",
  "mode": "auto",
  "polling_url": "/v1/tasks/tsk_..."
}
Calls to action

Choose the next honest surface.

Agent/developer

Review llms.txt for the verified staging path, eligibility boundary, and result-handling guidance.

Review llms.txt