# agentsite-static-version=e7701620-7d0c-4198-991d-a67b5a7f47ab
# levanto.ai

> AI-optimized mirror of levanto.ai containing 27 pages totalling 22,914 words of clean markdown content, structured data, and semantic HTML. Original source: https://levanto.ai. Last updated: 2026-11-09T12:30:43.837Z.

Each page is available as HTML (with JSON-LD structured data) and Markdown (text-only, ideal for LLMs and RAG).

## Homepage

- [Sage by Levanto Labs · A decision model API](/content/site-root.html): #1 of 27 on JevBench's official score (October 2026). Sage reads text and images (beta) and reasons when you turn it on. (190 words)

## Articles & Blog Posts

- [sagecraft/run-record-json.html](/content/sagecraft/run-record-json.html) (1,116 words)
- [docs/reproduce_agentharm-md.html](/content/docs/reproduce_agentharm-md.html) (2,343 words)
- [SageCraft — we put a dwarf in a harness and gave it a decision model | Levanto](/content/sagecraft/index.html): SageCraft is the open-source harness that lets Sage, Levanto’s decision model, play World of Warcraft. It reached level five, and Sage cost under three cents per kill. (1,198 words)
- [audit/decisions-2026-09-27-md.html](/content/audit/decisions-2026-09-27-md.html) (722 words)
- [Terms of Service | Levanto](/content/terms/index.html): Terms of Service for Levanto Labs, Inc. and the Sage inference API and related Services. (2,492 words)
- [mpp/levanto-sage-mpp-skill-md.html](/content/mpp/levanto-sage-mpp-skill-md.html) (416 words)
- [Privacy Policy | Levanto](/content/privacy/index.html): Privacy Policy for Levanto Labs, Inc. and the Sage inference service: what we collect, how long we retain it, and your rights. (987 words)
- [editorial/sage2-examples-json.html](/content/editorial/sage2-examples-json.html) (90 words)
- [Intelligence, at runtime | Levanto](/content/thesis/index.html): How decision models will eat code at runtime. Write-time already speaks English; runtime still does not. A thesis from Levanto Labs. (1,073 words)
- [Agentic security | Levanto](/content/agentic-security/index.html): Sage matches GPT-5 on AgentHarm at about 5× the speed: 77% of harm prevented, 95.5% of harmless requests allowed, 208 ms median. Reproduction guide included. (544 words)
- [Sage is #1 on JevBench | Levanto](/content/benchmarks/index.html): Sage scores 74.0, #1 on the JevBench composite (board v1.7.46, 14 Oct 2026). Plus Sage 1.4 vs Jev, and Sage Router on GPQA Diamond. (289 words)
- [Model routing | Levanto](/content/model-routing/index.html): Build a model router on Levanto Sage: frontier-grade judgment at classifier speed, with calibrated confidence on every answer. Head-to-head vs OpenRouter Auto Beta on AIME. (808 words)
- [Sage2 Decision and Extraction Models | Levanto Labs](/content/model/index.html): Sage2 extracts structured data and returns decisions with calibrated confidence. It supports streaming, reasoning, web grounding, and vision (beta). (526 words)
- [Sage and Jev pricing per 1,000 decisions | Levanto](/content/pricing/index.html): Sage costs about $0.025 per 1,000 decisions and Jev costs $0.032 (Benchmark Heaven, October 2026). Atlan measured Sage at $0.41 and Jev at $0.043. (635 words)
- [Sage on MPP: intelligence at runtime, no account | Levanto](/content/mpp/index.html): Your agent pays for Sage over MPP, by card or USDC, and spends it a call at a time. Decisions, extraction and streaming. No account. (225 words)
- [Case studies | Levanto](/content/case-study/index.html): How teams use Sage: the decisions, the integration and the results. (302 words)
- [Blog | Levanto](/content/blog/index.html): Product announcements, research and practical guides to decision models, confidential inference and building with Sage. (99 words)
- [Introducing Sage2: Extraction and Streaming Decisions | Levanto](/content/blog/introducing-sage2/index.html): Sage2 adds extraction and streaming decisions to Levanto’s runtime models. Extract typed fields from natural language and update decisions as context changes. (767 words, 2026-11-02)
- [Sage on NEAR AI: Confidential AI for Enterprise Data](/content/blog/sage-near-ai-confidential-enterprise-decisions/index.html): Sage launches on NEAR AI, bringing TEE-protected inference to large-scale data classification, sensitive data pipelines, and automated enterprise decisions. (1,101 words, 2026-10-26)
- [Decision Model Confidence: When to Remove “Uncertain”](/content/blog/decision-model-uncertain-option/index.html): Learn when to remove “uncertain” from decision model options, handle Sage null responses, and test confidence thresholds for automated workflows. (1,433 words, 2026-10-22)
- [Arena by Normies: Autonomous game agents powered by Sage](/content/case-study/arena-normies/index.html): Arena uses character context, persistent strategies, and structured decisions to let AI agents act independently in a shared survival game. (1,598 words, 2026-10-22)
- [Quirq tests lower-cost coding-agent routing with Sage](/content/case-study/quirq-coding-agent-routing/index.html): Quirq’s XO platform used Sage to select models, then retried unfinished work on a stronger model. Across 90 tasks, it reached similar final completion for $34.04 instead of $69.57. (1,093 words, 2026-10-08)
- [PerpFlow speeds up agentic research with Sage](/content/case-study/perpflow-agentic-research/index.html): Claude Code delegates thousands of market classifications to Sage, keeping bulk results in files while Python tests the findings against predefined criteria. (872 words, 2026-10-05)
- [Leap turns study requests into learning tools with Sage | Levanto](/content/case-study/leap-study-tool-routing/index.html): Leap routes requests for flashcards, quizzes, and other study activities from chat into native tools, with 89.9% accuracy in its reported evaluation. (707 words, 2026-10-01)
- [GrowBlocks turns monster descriptions into gameplay with Sage | Levanto](/content/case-study/growblocks-monster-descriptions-gameplay/index.html): Two authoring passes. 40 monster types. GrowBlocks brings natural-language personalities into combat decisions and adaptive dungeon encounters. (970 words, 2026-09-28)
- [Capacitr uses Sage to save money on market analysis | Levanto](/content/case-study/capacitr-market-signal-decision-gates/index.html): Six structured decisions help Capacitr filter market noise, focus research, and check its trading agent before expensive work begins. (318 words, 2026-09-21)

## Resources

- [Full Page Index](/index.html): Browse all cached pages with rich metadata
- [About This Cache](/about.html): Methodology, technical details, and usage guidelines
- [XML Sitemap](/sitemap.xml): Machine-readable sitemap for crawler discovery
- [Robots.txt](/robots.txt): Crawler directives
- [Agentsite Network](https://agentsite.network/network.html): Public index of Agentsites and their machine-readable resources