Jev in Practice: Browser Use and Agent Workflows

The author forked nanocode and added tools to grep a memory.md file and use Jev compaction to remove unnecessary memories.

The author built an awesome list for Jev projects where Jev checks source evidence, reviews descriptions and suggests categories, while maintainers decide what gets merged.

Jev only handles boolean checks, enum choices and candidate scoring, positioned as System One, with search intent routing, local model gateways and multi-agent dispatch as its main use cases; it is live on OpenRouter but needs the Decisions API.

Jevals replaces the LLM judge with Jev, so every pass/fail comes with a calibrated confidence you can set a threshold on, at lower cost and higher speed.

In a demo, Jev classified 100 emails in 1.42 seconds and handed the 31 with confidence below 95% to Kimi K3, with the two models together reaching 96% accuracy.

A joking teaser says someone will demo how Jev can speedrun getting into YC for 0.03 cents, part of community discussion of using Jev for agent tasks.

The author recommends the 139k-star awesome-llm-apps collection and links browser-use's jev-ultrafast repository.

A long German post explains TypeSafe AI's Jev: you preset options (low/medium/high risk) and it returns only probability numbers, with use cases sketched for risk control, customer service and compliance.

A demo uses Jev for real-time emotion detection, pairing each Bluesky post with an avatar matched to the user's profile, with a demo video attached.

A Monad developer built a purely AI-native on-chain trading bot with Jev, which outputs only classification probabilities instead of text, runs at millisecond latency against the Kuru order book, and is fully open-sourced; the 14-hour demo lost over 300%.

A developer asks whether anyone has used Jev as an LLM-as-judge evaluation model and invites people to share their approaches.

The author walks through applying for a Typesafe account, setting the TYPESAFE_API_KEY environment variable, and adding Jev to Pi Agent with the skills command.