Operated by Sourcelane✓ verified
Label, score and flag texts, images or scraped records in bulk with OpenAI GPT-6 Luna Decisions — your own categories, one typed answer and confidence per row.
A bulk multimodal classifier that uses OpenAI GPT-6 Luna Decisions to convert lists of texts, images, or scraped records into structured, typed answers. Users define custom decision questions (choice labels, ordered score levels, or yes/no checks); the connector returns one typed answer per question per item along with a confidence measure and per-option probabilities. It supports evaluating images alone or together with accompanying text, handles up to four images per item, truncates long text inputs, and answers multiple questions (batching up to ten) in a single pass using the Decisions API’s probabilistic, typed outputs rather than free-form text.
Verified Oct 25, 10:11 PM
$0.002
per evaluated item
Up to 100 per call. Only pay for results returned; failed calls are refunded.
Support ticket triage
Route messages to the correct team and surface urgent cases
Social media post classification
Label posts by caption and photo for type, category, or brand-safety
Product and listing image tagging
Categorize marketplace/e-commerce photos and flag low-quality or off-brand images
Review and comment analysis
Sentiment, topic, and complaint detection across many reviews
Lead qualification
Score scraped profiles or form submissions against custom ICP criteria
Content moderation
Flag spam, promotional, or unsafe text and images with tunable confidence thresholds
The texts to evaluate — one per line.
Images to evaluate — one URL per line, each evaluated on its own.
When reading from a results, only send these fields to the model, e.g.
When reading from a results, the fields that hold image URLs, e.g.
The questions to ask about every text.
Optional settings
| Window | Uptime | Success rate | p50 | p95 | Calls |
|---|---|---|---|---|---|
| 24h | — | — | — | — | 0 |
| 7d | — | — | — | — | 0 |
| 30d | — | — | — | — | 0 |
Call it through Sourcelane's gateway or MCP server. We run the connector, apply your agent's spend guardrails, and bill only the results returned.
Call it via Sourcelane
curl -X POST https://api.usesourcelane.com/v1/call \
-H "Authorization: Bearer sl_live_your_agent_key" \
-H "Content-Type: application/json" \
-d '{"listing":"gpt-bulk-classifier","params":{"questions":[{"key":"department","type":"choice","instructions":"Which team should handle this message?","criteria":{"billing":"Payment or subscription issues","technical":"Bugs or integration problems","sales":"Pricing or account questions"}},{"key":"frustration","type":"score","instructions":"How frustrated the customer appears","criteria":["Calm, just stating facts","Frustrated but civil","Very angry, strong language"]},{"key":"is_urgent","type":"noul","instructions":"The message conveys urgency or time-sensitivity"}],"maxItems":5},"maxResults":10}'Request params
| Field | Type | Required | Description |
|---|---|---|---|
texts | array | optional | The texts to evaluate — one per line. Leave empty if you are reading from an existing results instead. |
imageUrls | array | optional | Images to evaluate — one URL per line, each evaluated on its own. JPEG, PNG, WebP and GIF work; large images are resized automatically. |
fields | array | optional | When reading from a results, only send these fields to the model, e.g. 'text' and 'title'. Leave empty to send the whole record. |
imageFields | array | optional | When reading from a results, the fields that hold image URLs (a URL or a list of URLs), e.g. 'displayUrl' or 'images'. Those images are evaluated together with the record's text. |
questions | array | required | The questions to ask about every text. Each question has a 'key' (the output column name), a 'type' ('choice' picks one option, 'score' rates against ordered levels, 'noul' answers yes/no), 'instructions', and 'criteria' (options for choice, ordered levels for score). |
maxItems | integer | optional | Maximum number of texts to evaluate in this run. Set 0 for no limit. Capped at 100 per call. |
concurrency | integer | optional | How many texts to evaluate in parallel. Higher is faster; lower it if you hit rate limits. |
maxImagesPerItem | integer | optional | When a results record has several images, evaluate at most this many of them (1–4). |
Each result contains
| Field | Type | Description |
|---|---|---|
itemIndex | integer | Item |
text | string | Text |
shows_people | boolean | — |
shows_people_probability | number | — |
topic | string | — |
topic_confidence | number | — |
topic_probabilities | object | — |
Pick your tool and connect in under a minute. Then just ask — for example: “Transcribe this video and give me timestamps for every product mention: <url>”
Connect Claude
Web, desktop and mobile. One click, then confirm.
Adds the Sourcelane MCP server with your key as a header.
claude mcp add --transport http sourcelane https://api.usesourcelane.com/mcp \
--header "Authorization: Bearer sl_live_your_agent_key"Alternative to the connector URL: add to claude_desktop_config.json, then restart Claude.
{
"mcpServers": {
"sourcelane": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://api.usesourcelane.com/mcp",
"--header",
"Authorization:${AUTH_HEADER}"
],
"env": {
"AUTH_HEADER": "Bearer sl_live_your_agent_key"
}
}
}
}Add to ~/.cursor/mcp.json (or Windsurf's mcp_config.json).
{
"mcpServers": {
"sourcelane": {
"url": "https://api.usesourcelane.com/mcp",
"headers": {
"Authorization": "Bearer sl_live_your_agent_key"
}
}
}
}Save as .vscode/mcp.json. VS Code prompts for your key once.
{
"servers": {
"sourcelane": {
"type": "http",
"url": "https://api.usesourcelane.com/mcp",
"headers": {
"Authorization": "Bearer ${input:sourcelane-key}"
}
}
},
"inputs": [
{
"type": "promptString",
"id": "sourcelane-key",
"description": "Sourcelane agent key",
"password": true
}
]
}Create a GPT → Actions → Import from URL, then Authentication: API Key, Bearer.
https://usesourcelane.com/openapi.jsonOne POST. Pass maxResults to cap cost.
curl -X POST https://api.usesourcelane.com/v1/call \
-H "Authorization: Bearer sl_live_your_agent_key" \
-H "Content-Type: application/json" \
-d '{"listing":"gpt-bulk-classifier","params":{"questions":[{"key":"department","type":"choice","instructions":"Which team should handle this message?","criteria":{"billing":"Payment or subscription issues","technical":"Bugs or integration problems","sales":"Pricing or account questions"}},{"key":"frustration","type":"score","instructions":"How frustrated the customer appears","criteria":["Calm, just stating facts","Frustrated but civil","Very angry, strong language"]},{"key":"is_urgent","type":"noul","instructions":"The message conveys urgency or time-sensitivity"}],"maxItems":5},"maxResults":10}'Wrap the REST call as a tool, or point an MCP client at the endpoint.
import requests
res = requests.post(
"https://api.usesourcelane.com/v1/call",
headers={"Authorization": "Bearer sl_live_your_agent_key"},
json={
"listing": "gpt-bulk-classifier",
"params": {"questions":[{"key":"department","type":"choice","instructions":"Which team should handle this message?","criteria":{"billing":"Payment or subscription issues","technical":"Bugs or integration problems","sales":"Pricing or account questions"}},{"key":"frustration","type":"score","instructions":"How frustrated the customer appears","criteria":["Calm, just stating facts","Frustrated but civil","Very angry, strong language"]},{"key":"is_urgent","type":"noul","instructions":"The message conveys urgency or time-sensitivity"}],"maxItems":5},
"maxResults": 10,
},
timeout=300,
)
body = res.json()
print(body["receipt"]["chargedMicros"], "micro-USD for", body["receipt"]["results"], "results")
print(body["data"])—
0 reviews
Leave a review
Loading reviews…