# HUMANT

> HUMANT is a founder-led beta building an exception runtime for AI agents. The current live product wedge is voice-call execution for eligible tasks. AI attempts the call first; a human may escalate when needed; the calling workflow receives a structured result.

## Machine-readable status

```yaml
company: HUMANT
status: beta
production_deployment_claim: false
current_live_wedge: voice-call execution
selected_customer_vertical: none_selected
staging_base_url: https://staging.humant.ai
create_task: POST /v1/dispatch
list_tasks: GET /v1/tasks
get_task: GET /v1/tasks/{task_id}
authentication: Bearer API key issued to an authorized beta account
api_documentation: /docs/
openapi: /openapi.json
```

## Should an agent use HUMANT?

Use only if all are true:

1. The task is a bounded voice-call objective.
2. The caller has HUMANT beta authorization.
3. The call is lawful, non-emergency, consent-aware, and allowed by HUMANT policy.
4. The workflow can accept an honest failure, review state, or unknown outcome rather than fabricating completion.

Do not use HUMANT for regulated advice, emergency services, deception, spam, harassment, credential collection, CAPTCHA/paywall/access-control bypass, physical-world verification, broad research, form completion, or any non-voice task type unless HUMANT later verifies that capability.

## Create a voice-call task

```http
POST https://staging.humant.ai/v1/dispatch
Authorization: Bearer <beta_api_key>
Content-Type: application/json
```

```json
{
  "type": "voice_call",
  "phone_number": "+14155550123",
  "brief": "Confirm whether Thursday at 2pm is available",
  "callback_url": "https://your-agent.example/humant-result"
}
```

`callback_url` is optional and must use HTTPS. Without it, poll the returned task path.

## Initial response

A successful dispatch returns HTTP `202` with fields including:

```json
{
  "task_id": "tsk_...",
  "status": "queued",
  "type": "voice_call",
  "mode": "auto",
  "polling_url": "/v1/tasks/tsk_..."
}
```

Treat all examples as staging beta interface documentation, not a performance, completion, availability, or service-level guarantee.

## Result handling

Poll `GET /v1/tasks/{task_id}` with the same bearer credential or receive the callback where configured. Task states in the current implementation include queued, screening, dispatched, in progress, completed, and failed paths.

Treat HUMANT results as bounded operational evidence, not ground truth. Continue the parent workflow through its normal exception path when a task fails, requires further review, or returns insufficient evidence. Never invent completion.

Recording, transcription, summaries, and other artifacts are conditional on task outcome, consent, jurisdiction, policy, and availability. Their presence is not guaranteed.

## Human escalation

Human escalation is possible but not guaranteed. Do not infer an on-demand workforce, response-time average, worker earnings, task volume, or service-level agreement.

## Payments and pricing

No current public price, prepaid balance, x402 settlement path, autonomous payment flow, or worker payout claim is approved on this staging page.

## Discovery and next actions

- Human or investor: email `hello@humant.ai` to request a founder conversation.
- Developer: request beta access at `/login?tab=register`.
- API documentation: `/docs/`.
- OpenAPI document: `/openapi.json`.

The public website and this file deliberately describe only the approved voice-call beta wedge, even where the repository contains broader experimental code paths.