QA Agent

QA Agent is in beta. The request and response format can change.

QA Agent researches a record (a lead, a seller, a company) against criteria you write in plain English and returns an assessment in notes plus the fields you define in struct, such as a score or a YES/NO verdict. Use it for lead qualification, marketplace vetting, and CRM hygiene.

Endpoint

POST https://api.sixtyfour.ai/qa-agent
POST https://api.sixtyfour.ai/qa-agent-async

API Reference

See the full request/response schema and parameters in the API Reference.

Request

Parameter	Type	Required	Description
`data`	object	Yes	The record to evaluate. Any keys and values. Must not be empty.
`qualification_criteria`	array	Yes	The criteria to evaluate against. Must contain at least one item.
`struct`	object	No	Fields to return in `structured_data`, as `field_name: "description"` pairs.
`references`	array	No	Pages the agent reads before it starts researching.
`max_tool_calls`	integer	No	Maximum research actions per evaluation, 1–50. Default `10`.
`webhook_url`	string	No	HTTPS URL that receives the result when the job completes. Sync calls send it too. See Outgoing Webhooks.
`dispatch_id`	string (8–128 chars)	No	Async only. Idempotency key: a repeat returns the first job and charges nothing. See Async Jobs.

`qualification_criteria`

Field	Type	Required	Description
`criteria_name`	string	Yes	Short name for the criterion.
`description`	string	Yes	What qualifies a record, in plain English.
`weight`	number	No	Relative importance of this criterion. Default `1.0`.
`threshold`	number	No	The minimum the record must meet on this criterion. It has no fixed scale; give the scale in `description`.

The agent reads every criterion, including its weight and threshold, as guidance for its research and assessment. QA Agent does not compute a score itself: to get one, ask for it in struct and state the scale you want (for example "Weighted qualification score from 0 to 10").

`references`

Each reference is an object with a required url and an optional description of what the page shows.

{
  "references": [
    {"url": "https://www.linkedin.com/company/sixtyfourai", "description": "Company profile"}
  ]
}

`struct`

Each key in struct becomes a field in structured_data, and its value is a plain-English description of what to put there. Only the fields you list are returned. Specific descriptions work best: "Overall qualification score from 0 to 10, weighted by criterion" beats "score". struct values must be strings; typed objects such as {"type": "int"} are rejected. Every returned value is a string, and an unfilled field is an empty string. Without a struct, structured_data contains a single general_findings field.

Response

Field	Type	Description
`structured_data`	object	One string value per `struct` field.
`notes`	string	The agent’s assessment: findings, how the record measures up against each criterion, and gaps in the evidence.
`references`	object	Always empty; sources are cited in `notes`.

{
  "notes": "Alex Johnson runs an online store selling handmade ceramics and ships from their own studio. Physical Inventory: met; product photos and shipping policy show stock held on site...",
  "structured_data": {
    "suitability_score": "9",
    "final_verdict": "YES"
  },
  "references": {}
}

Sync usage

curl -X POST "https://api.sixtyfour.ai/qa-agent" \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "data": {
      "name": "Alex Johnson",
      "country": "United Kingdom",
      "business_type": "e-commerce seller",
      "website": "https://example.com"
    },
    "qualification_criteria": [
      {
        "criteria_name": "Physical Inventory",
        "description": "Must own and store physical products",
        "weight": 10.0
      }
    ],
    "struct": {
      "suitability_score": "Overall qualification score from 0 to 10",
      "final_verdict": "YES or NO"
    }
  }'

import requests

response = requests.post(
    "https://api.sixtyfour.ai/qa-agent",
    headers={"x-api-key": "YOUR_API_KEY", "Content-Type": "application/json"},
    json={
        "data": {
            "name": "Alex Johnson",
            "country": "United Kingdom",
            "business_type": "e-commerce seller",
            "website": "https://example.com",
        },
        "qualification_criteria": [
            {
                "criteria_name": "Physical Inventory",
                "description": "Must own and store physical products",
                "weight": 10.0,
            }
        ],
        "struct": {
            "suitability_score": "Overall qualification score from 0 to 10",
            "final_verdict": "YES or NO",
        },
    },
)
response.raise_for_status()
result = response.json()
print(result["structured_data"]["final_verdict"])

const response = await fetch("https://api.sixtyfour.ai/qa-agent", {
  method: "POST",
  headers: { "x-api-key": "YOUR_API_KEY", "Content-Type": "application/json" },
  body: JSON.stringify({
    data: {
      name: "Alex Johnson",
      country: "United Kingdom",
      business_type: "e-commerce seller",
      website: "https://example.com"
    },
    qualification_criteria: [
      {
        criteria_name: "Physical Inventory",
        description: "Must own and store physical products",
        weight: 10.0
      }
    ],
    struct: {
      suitability_score: "Overall qualification score from 0 to 10",
      final_verdict: "YES or NO"
    }
  })
});

const result = await response.json();
console.log(result.structured_data.final_verdict);

Async pattern

Use /qa-agent-async for batches and anything that runs inside a request timeout.

Submit: POST /qa-agent-async with the same body as the sync endpoint. The response is {"task_id": "...", "status": "RUNNING"}.
Poll: GET /job-status/{task_id} every 10 seconds until status is no longer running.
Read: when completed, the response’s result holds the same object the sync endpoint returns.

Pass webhook_url to receive the result instead of polling. Async Jobs covers statuses, the polling budget and retries.

A sync call keeps running after your client times out or disconnects, and it is charged when it finishes. A retry starts and charges a second run. To retry safely, use /qa-agent-async with dispatch_id.

curl -X POST "https://api.sixtyfour.ai/qa-agent-async" \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "data": {"name": "Alex Johnson", "business_type": "e-commerce seller"},
    "qualification_criteria": [
      {"criteria_name": "Physical Inventory", "description": "Must own and store physical products"}
    ],
    "struct": {"final_verdict": "YES or NO"}
  }'

curl "https://api.sixtyfour.ai/job-status/TASK_ID" \
  -H "x-api-key: YOUR_API_KEY"

import time
import requests

BASE_URL = "https://api.sixtyfour.ai"
headers = {"x-api-key": "YOUR_API_KEY", "Content-Type": "application/json"}

start = requests.post(
    f"{BASE_URL}/qa-agent-async",
    headers=headers,
    json={
        "data": {"name": "Alex Johnson", "business_type": "e-commerce seller"},
        "qualification_criteria": [
            {"criteria_name": "Physical Inventory", "description": "Must own and store physical products"}
        ],
        "struct": {"final_verdict": "YES or NO"},
    },
)
start.raise_for_status()
task_id = start.json()["task_id"]

while True:
    response = requests.get(f"{BASE_URL}/job-status/{task_id}", headers=headers)
    if response.status_code in (429, 503):
        time.sleep(float(response.headers.get("Retry-After", 10)))
        continue
    response.raise_for_status()
    job = response.json()
    if job["status"] == "completed":
        print(job["result"]["structured_data"])
        break
    if job["status"] != "running":
        raise RuntimeError(f"QA Agent job {job['status']}")
    time.sleep(10)

const BASE_URL = "https://api.sixtyfour.ai";
const headers = { "x-api-key": "YOUR_API_KEY", "Content-Type": "application/json" };

const start = await fetch(`${BASE_URL}/qa-agent-async`, {
  method: "POST",
  headers,
  body: JSON.stringify({
    data: { name: "Alex Johnson", business_type: "e-commerce seller" },
    qualification_criteria: [
      { criteria_name: "Physical Inventory", description: "Must own and store physical products" }
    ],
    struct: { final_verdict: "YES or NO" }
  })
});
const { task_id } = await start.json();

while (true) {
  const response = await fetch(`${BASE_URL}/job-status/${task_id}`, { headers });
  if (response.status === 429 || response.status === 503) {
    const retryAfter = Number(response.headers.get("Retry-After") ?? 10);
    await new Promise((r) => setTimeout(r, retryAfter * 1000));
    continue;
  }
  if (!response.ok) throw new Error(`Job status ${response.status}: ${await response.text()}`);
  const job = await response.json();
  if (job.status === "completed") {
    console.log(job.result.structured_data);
    break;
  }
  if (job.status !== "running") {
    throw new Error(`QA Agent job ${job.status}`);
  }
  await new Promise((r) => setTimeout(r, 10000));
}

Pricing

See Credit Costs for more information.

Errors

For error responses (400, 403, etc.), see Handling Errors.

Get Started

Intelligence API

Atlas

Lattice

Search API

Workflows API

Monitors API

Webhooks

Developer Tools

Endpoint

API Reference

Request

`qualification_criteria`

`references`

`struct`

Response

Sync usage

Async pattern

Pricing

Errors

​Endpoint

API Reference

​Request

​qualification_criteria

​references

​struct

​Response

​Sync usage

​Async pattern

​Pricing

​Errors

Endpoint

Request

`qualification_criteria`

`references`

`struct`

Response

Sync usage

Async pattern

Pricing

Errors