For Agents & AI teams

Give your agent hands in China.

When a task needs local supplier communication, physical verification, human judgment or accountable follow-through, define a bounded handoff to SeekAPI. The current public handoff is a manual-review preview; no live human-task API is claimed.

Handoff trigger

Use a human when the task crosses the software boundary.

Software can research, compare and plan. Human execution becomes relevant when the work requires Chinese-language contact, a physical observation, commercial authority, judgment or persistence across several people.

Proposed task classes

Eight task types, each subject to scope review.

These classes define the service vocabulary. They do not claim universal availability, location coverage or automatic acceptance.

AH01

Supplier contact

Contact a named supplier and return structured answers without changing commercial terms.

AH02

Discovery & qualification

Find candidates, check available evidence and return capability gaps and unknowns.

AH03

RFQ execution

Send an approved revision to approved recipients and normalize responses.

AH04

Sample / NPI follow-up

Track sample status, revision, issues and buyer approval evidence.

AH05

Physical verification coordination

Coordinate observable evidence or qualified third-party inspection where available.

AH06

Quality / exception recovery

Follow up on delay, rework, nonconformance or loss of communication.

AH07

Delivery, logistics & warehousing

Coordinate packaging, freight, shipment milestones, warehouse evidence, inventory status and local readiness within the agreed role.

AH08

Compliance research + human verification

Add human verification to entity, license, product-path, document, contact or other critical business facts.

Illustrative handoff

State what humans may do, must return and must never infer.

Contact authority does not imply RFQ, spending, contract or production authority. Sensitive files are excluded from this preview.

requester_type: agent
organization: example_ai_team
objective: confirm_sample_revision_status
task_type: AH01
context_summary: illustrative_only
allowed_actions: [contact_supplier, request_status]
prohibited_actions: [change_terms, make_payment, accept_deviation]
recipient_permissions: [named_supplier]
max_budget: null
deadline: null
required_evidence: [supplier_response, photo_if_available]
completion_definition: status_and_open_questions_returned
human_approval_contact: required
Illustrative evidence return

“Done” is not a sufficient result.

A completed or blocked task should return actions, sources, limitations, costs, unknowns and the next decision.

status: COMPLETED | PARTIAL | BLOCKED | DECLINED
actions_taken: [{actor_type, action, timestamp}]
findings: []
evidence: [{source, created_by, timestamp, limitation}]
costs: {approved: 0, actual: 0, third_party: 0}
open_questions: []
exceptions: []
next_decision: null
Agent Human Desk

Reserved human execution capacity under agreed permissions.

For AI teams with recurring China-side work, a Desk can define task classes, concurrent work, response windows, evidence, escalation contacts and spending authority. Scope and service levels require agreement; this preview is not a live subscription.

Current state

Manual review preview.

The task vocabulary, permissions and evidence format are defined. Manual scope questions can be sent by email, but structured web intake, universal coverage, public task API, MCP and WebMCP execution are not live on this preview.

Common questions

Automation does not bypass authority.

Is a public human-task API live?

No. The current public route is manual scope review. A task API, MCP and WebMCP execution are not live on this preview.

How does an Agent start?

The Agent owner or developer prepares a non-sensitive task class, objective, allowed actions, prohibited actions, evidence requirements and human approval contact for manual review.

Can an Agent authorize payment or accept a quality deviation?

Not by default. Payment, contract changes, quality deviations and other high-risk actions require explicit human authority. No payment is available in this preview.

Prepare a bounded handoff.

Validate a non-sensitive draft in the browser. Nothing is submitted or accepted in this preview.