jevcrawl
Choose crawl links by goal and return relevant source passages
Implementation and evidence →JEV GUIDE
Reviewed against official documentation on October 15, 2026. Model availability, SDK releases and community-project compatibility are separate questions. We have not benchmarked the model.
Jev 1.14 became available October 14. jev-preview points to jev-1.14.0; jev-latest still points to jev-1.13.0. Both cost $0.042 per million input tokens, with free output. Pin versions for reproducibility and record the response model.
Source: TypeSafe · Models ↗
Images are off by default. Both 1.13 and 1.14 require TypeSafe-Danger-Zone-Images: true. Assets can supply images in state, instructions or criteria. TypeSafe describes them as dreamer-ready, not production-ready; formats, limits and accuracy may change.
Source: TypeSafe · Images ↗
Box requires jev-1.14.0 or jev-preview. Each question returns one rectangle with normalized top, bottom, left and right edges plus confidence. It has no not-found result. Pair it with a Noul presence check before using its coordinates.
Source: TypeSafe · Box ↗
Defaults are 64k per request and 32k per question. On both versions, TypeSafe-Danger-Zone-Long-Context: true raises both to 262k. The documented token price is unchanged, but processing is slower and less accurate.
Source: TypeSafe · Models ↗
Released October 14, version 0.7.0 requires Node.js 22+, stops exporting request payload types, and adds images and Box. On Node.js 22.15+, bodies of at least 1 KiB use zstd by default; compression: false disables it. Check runtime and type imports first.
Python 0.7.5 adds Image, Box and attach_assets, with zstd request compression enabled by default. Its changelog warns that unofficial APIs may not support compression immediately. Validate SDK, model and gateway changes separately.
Run the same business samples against old and new configurations, record errors, fallback behavior, latency and cost, then recalibrate thresholds. Newly curated projects are not automatically verified for 1.14; their pages retain the model and dependency boundaries of the reviewed source.
Choose crawl links by goal and return relevant source passages
Implementation and evidence →Gate command-line text and downstream commands with a Jev probability
Implementation and evidence →Select OpenCode agents and decide when to hand work over
Implementation and evidence →02 / JEV GUIDE
Understand TypeSafe Jev, its typed outputs and practical use cases, with browser automation, MCP and context-filtering projects.
Read guide →03 / JEV GUIDE
Build a Jev API request with state, typed questions and Bearer authentication. Download a JSON fixture and review response checks.
Read guide →15 / JEV GUIDE
Compare Jev and generative LLM task boundaries through rewriting and script-review cases. Typed output does not establish correctness or performance.
Read guide →