Sage by Levanto: AI Text Classification API

What is Levanto Sage used for?

Sage is an AI text classification and decision API from Levanto Labs. Think of it as a sorting desk built into your software. If a customer writes “I was charged twice,” Sage can label the message as a billing issue. Your application can then send it to the right queue. The same API can pull individual fields out of a message or update a decision as more of a conversation arrives.

Pick the right kind of decision

Use Choice when a ticket should land in one of several queues, such as billing, shipping, or account access. Use Tags when several labels can apply to the same message. Yes/No checks a specific condition, Scale scores against your rubric, and Sort orders items by a criterion you supply. Start with the result your application can act on. For example, deciding whether a ticket needs review is easier to test than asking the model to explain the entire case.

Build and test one routing rule

1. Create a key under Decision model → API keys in the Sage dashboard. Keep it on your server and send it as a Bearer credential.
2. Send a sample message to POST /decide. Supply the text as content and a yesno question with an ID and instructions such as “Does this message report a duplicate charge?”
3. Inspect result.answer and compare it with your expected label. Try examples that should return yes and no, then give uncertain results and request errors their own review path before letting the model route live traffic.

Extract fields or update a decision live

Call POST /v1/extract with text and field definitions to collect a stated amount, date, or reference from a message. Check the matched span when the value matters and handle missing fields explicitly. For a chat that is still in progress, open a WebSocket session at /v1/stream, define the question, and append incoming text. Earlier text cannot be edited in place, so start a fresh session if a transcript needs correcting.

Cost and data handling

The public USD schedule checked November 10, 2026 includes a monthly free allowance and paid monthly plans. Input tokens, output tokens and any searches that run count against usage. Estimate a decision before sending it with POST /usage/estimate, and inspect actual usage after the call. Current Sage pricing

FAQ

Can I try Sage for free?

Yes. The Free plan includes a monthly allowance. After it runs out, calls return HTTP 402 until the next period or an upgrade. Unused allowance expires.

Does Levanto train on my API data?

Levanto says it does not train models on Sage inputs or outputs. Ordinary inference content is discarded after the response. Failed or unusually slow requests can have diagnostic payloads kept for 30 days. Levanto privacy policy

What if the model is unsure?

A Yes/No result can return null. Give it a separate path, such as human review. You can also check the reported probability against a threshold tested on your own examples.

Official developer resources

• Sage quickstart — Make an authenticated request and inspect a complete response.
• Levanto Labs on GitHub — Find public examples and benchmark code.