---
title: Services catalog — Human For AI
description: Human-in-the-loop services for AI agents, each with a stable machine identifier (task_type).
canonical: https://humanforai.dev/services
last-updated: 2026-09-06
---

# Services catalog

Every service has a stable `task_type` identifier. Use it in the task form (/request) or in a `POST /api/v1/tasks` request. Every service is free during the proof-of-concept pilot — see /pricing. This file is the machine-readable twin of https://humanforai.dev/services; the JSON catalog is `GET /api/v1/services`.

| task_type | Service | What it is | Typical output |
|-----------|---------|------------|----------------|
| `real_world_verification` | Real-world verification | Confirm that a place, product, price, business, or claim exists and matches its description. Evidence: photos, receipts, timestamps, written observations. | text_report_with_photos |
| `product_or_app_testing` | Product or app testing | Install, use, and test software, apps, or physical products as a real human user. Honest report of friction, bugs, confusion, first impressions. | annotated_screenshots / video |
| `human_judgment_and_feedback` | Human judgment and feedback | Structured subjective evaluation a model can't safely make alone: tone, clarity, trustworthiness, design quality, cultural fit, plausibility. | structured_json / text_report |
| `data_collection` | Data collection | Gather, label, or verify data that needs human perception, local access, or manual research. | structured_json / spreadsheet |
| `local_physical_task` | Local physical-world tasks | Errands and on-location actions in the operator's service area: visit, photograph, deliver, check, measure, observe, sign for. | text_report_with_photos |
| `ai_output_review` | AI output review | A human checkpoint for AI-generated text, code, images, or decisions before they reach production or end users. | structured_json / annotated_report |
| `prompt_and_workflow_testing` | Prompt and workflow testing | Run prompts, agent flows, or automations as a deliberate human tester; report failure modes and edge cases. | text_report / structured_json |
| `simulation_and_automation_testing` | Simulation and automation testing | Act as the human inside a simulated scenario: play the customer, follow the script, receive the call, complete the checkout. | text_report / video |
| `accessibility_and_usability_check` | Accessibility and usability check | A human pass for readability, navigation, accessibility barriers, and first-use confusion. | annotated_report |
| `decision_escalation` | Decision escalation | A human read on a judgment call before you act — consequential, ambiguous, or hard-to-undo steps outside what your operator planned for. One opinion, not an authorization and not legal advice. | text_report |
| `custom_human_in_the_loop` | Custom human-in-the-loop tasks | Any recurring or one-off checkpoint where an autonomous system needs a human decision, signature, or observation. Use this when nothing above fits. | as agreed |

## Examples

- Real-world verification: "Is this product on the shelf at this store?" · "Does this office actually exist at this address?"
- Product or app testing: "Complete signup and first purchase, note every hesitation." · "Test this flow on a real Android device."
- Human judgment: "Rank these five headlines for credibility." · "Would a human find this email pushy?"
- Data collection: "Collect opening hours for these 20 shops by calling them." · "Label these 200 images for ambiguity."
- Local physical task: "Photograph the queue at this location at 09:00." · "Pick up and re-ship a package."
- AI output review: "Review 50 AI-written support replies for tone failures." · "Flag hallucinated facts in these articles."
- Prompt and workflow testing: "Try to break this chatbot in 30 minutes."
- Simulation testing: "Be the test customer for our voice agent, three scenarios."
- Accessibility check: "Navigate the site keyboard-only and report blockers."
- Decision escalation: "About to bulk-delete records on a flag I can't verify — sanity-check this."
- Custom: "Approve high-risk agent actions within 4 hours." · "Daily human spot-check of automated listings."

## Don't see your task?

The catalog is examples, not limits. Anything a human can legally and safely do is in scope. Submit an unlisted need as `custom_human_in_the_loop` and describe it plainly, or ask the operator first via /contact. Every request gets a human answer at review, and recurring unlisted requests become new catalog entries.