Sage2 by Levanto Labs: Turn live talk into structured data and instant actions
Levanto Labs released Sage2 on November 2, 2026, as an update to its Sage decision model API, according to the company's launch post. The release adds two capabilities. The first is extraction: developers define the fields they need and assign each a type, such as text, number, money, date, time or duration. The API returns typed values, the passage each value came from, and a probability for each one. Fields the input does not mention come back as null. In Levanto's example, a message asking for a midsize SUV for up to $450 a month returns a vehicle, a $450 monthly limit, and no loan term.
The second capability is streaming decisions. A client opens a WebSocket session, appends text as a conversation grows, and receives an updated answer when a result changes, such as a question moving from no to yes. A third-party write-up puts session limits at 32 questions and 16,384 tokens. Levanto says the server keeps the conversation and reuses cached computation, so clients send only new text, and that streaming runs three times faster while using fewer tokens. Reasoning, web grounding and vision remain in beta. Product Hunt carried the listing with 210 upvotes and 3 comments, while a separate digest counted 201 upvotes.
Why it matters
For founders building voice agents, support tools or intake workflows, the release targets a specific cost: turning spoken or typed input into fields and routing decisions on every turn without prompting a general chat model each time. Levanto frames Sage as a narrow decision layer that returns null when unsure, so teams can send uncertain cases to a person. That makes confidence thresholds and handoff rules part of the product design rather than an afterthought.
For investors and buyers, the pitch rests on cost and latency. Levanto's launch materials claim the model is far cheaper than a named frontier model and beats a Cloudflare extraction model, but these are vendor figures. Buyers should test accuracy and cost on their own data. Per-session limits and a hosted, server-side state model also affect unit economics, so pricing and rate limits matter as much as the headline benchmarks.
Key points
- 1Levanto Labs launched Sage2 on November 2, 2026, adding extraction and streaming decision APIs to its Sage product.
- 2Extraction covers text, numbers, money, dates, times and durations, each returned with source text and a probability.
- 3Streaming sessions run over WebSocket and push a new answer when a tracked decision changes.
- 4A third-party write-up reports session limits of 32 questions and 16,384 tokens per streaming session.
- 5
