jev-ultrafast
Jev selects browser operations and elements; a separate text model fills text fields.
TypeSafe API key, text-model provider and a browser.
Repository & README ↗JEV / COMMUNITY PROJECTSOFFICIAL / TYPESAFE browser-use Ying-Kai-Liao awlevin droidrun moritzkremb tamaratran y0usaf kerpopule gargpratyush devagrawal09 kitze superagents-lab itsmostafa TianyuCodings ikermoel
See what people are building with Jev.
From browser actions to coding agents: find implementations worth exploring by task. Summaries and setup notes are based on project READMEs, not a shared performance benchmark.
Sources checked: 2026-09-20 · Community projects have not been individually installed or run; status may change.
Start with the official agent skill
The official TypeSafe skill helps your agent choose Choice, Score and Noul, read current docs and compose judgments correctly. It provides development context; it does not create API credentials or grant model access.
npx skills add typesafe-ai/skills --skill typesafe-aiFor Codex and other agents, select your agent when prompted. Claude Code users can use the official plugin installation method. Choose one installation method.
Jev AI getting-started guide →Browser & device automation
jev-browser
An LLM plans each step while Jev selects actions from a page snapshot. Available as a library, CLI and MCP server.
Your own TypeSafe key and Playwright; review action permissions.
Repository & README ↗typesafe-computer-use
A macOS control loop turns accessibility information and OCR into choices for Jev.
macOS permissions and a TypeSafe key; text generation may use another model.
Repository & README ↗mobile-jev
An Android agent with a CLI, live studio and execution traces, built on Mobilerun.
A Mobilerun device, Mobilerun API key and TypeSafe key.
Repository & README ↗jev-voice-browser
Speech recognition supplies text; Jev selects browser intent and targets while code decides when to act.
Microphone permission, Web Speech support, Playwright and a TypeSafe key.
Repository & README ↗Coding agents & workflows
fast-jev-compaction
Scores tool calls and results to remove stale context while keeping retained content verbatim.
Claude Code plugin or npm library; evaluate information loss on your own tasks.
Repository & README ↗pi-jev
Adds a tool-call gate, command-output checks and a typed judgment tool to the Pi coding agent.
Pi and a model API key; model judgments do not replace permission controls.
Repository & README ↗hermes-jev-skills
A collection for model routing, memory selection, context selection and bounded agent actions.
Check each skill’s supported host and provider configuration.
Repository & README ↗jev-router
Chooses a model tier for each fresh Claude Code or Codex user turn using Jev judgments.
An installed, authenticated CLI and a Jev provider key.
Repository & README ↗jev-review
Reviews diffs or codebases through focused risk, evidence and severity judgments, with a local dashboard.
A TypeSafe key; findings still need source review and tests.
Repository & README ↗skillbox
A self-hosted, versioned skill library with MCP access and optional Jev recommendations.
Jev is optional; recommendations use your selected provider and key.
Repository & README ↗Search & integrations
jev-search
Jev chooses sources, search terms and relevance rankings around Search1API results. Returns links and snippets.
A search service and a supported Jev provider; not a generated-answer engine.
Repository & README ↗typesafe-mcp
Exposes typed Jev evaluations to MCP clients through the evaluate command.
A compatible client and provider key; this is a community integration.
Repository & README ↗Independent implementations
NanoJev
A small parallel decision-model replica with its own training pipeline, model and game demos.
Independent research model, not TypeSafe’s Jev weights or an official release.
Repository & README ↗open-alternative-jev
An independent System One-style approach that produces constrained decisions from open-weight language models.
Self-hosted model/GPU requirements vary; not the official Jev model.
Repository & README ↗