Jev ecosystem projects — page 2

Give jevos text or structured state and a yes/no question; it returns the probability of yes. The local server uses Jev's wire format for Noul requests, allowing existing clients to address a separate self-hosted model.

typesafe-computer-use turns a written goal into desktop actions using OCR and accessibility controls on a Mac. TypeSafe Jev chooses the action, target and site; code supplies facts, while a separate writer handles free text. Provide a TypeSafe key and screen-recording and accessibility permissions. The CLI offers a dry run before real mouse and keyboard control.

RAG-Fusion generates query variants and merges retrieval rankings with Reciprocal Rank Fusion to find material a single query may miss. Jev is an optional reranker in the evaluation pipeline; its experiments require JEV_API_KEY. The original demo remains vector-only. Compare retrieval strategies on NFCorpus with the evaluation harness; the hybrid pipeline combines BM25, vector search, fusion and reranking.

Reticle checks running apps using network, console and state evidence, returning pass, fail or unknown. The optional Jev exploration driver selects DOM candidates; internal verification-flow routing remains planned. Connect a development app through MCP, describe expected behavior and save flows for deterministic replay.

Hermes Jev Skills provides reusable skills for model routing, search, memory filtering, triage and browser actions. Jev makes bounded choices such as selecting a model or ranking retrieved passages; the host model handles prose. The skills work with Hermes, Claude Code and Codex. Shadow mode records routing decisions without switching models.

Awesome Jev is heyjunpenn’s community catalog of Jev projects, organized into SDKs, agents, applications, research and other categories. Entries explain what Jev decides and track implementation evidence, reproduction status and metric provenance. Browse by category, then use the evidence labels and upstream links to inspect a project’s actual decision workflow.

decider returns Choice, Score and Noul probabilities from a state and typed questions in one forward pass. This independent Jev-style model family uses Qwen bases; its training data was not distilled from Jev. Call the Python Decider interface or point TypeSafe SDKs at its compatible local HTTP endpoint.

ollaya downloads named open decision models and serves typed questions through a local daemon. It implements TypeSafe-compatible endpoints for existing Jev clients, while remaining an independent project. Use its CLI, desktop app or MCP server; Modelfiles can package reusable question sets.

AnyJev reads probabilities from open models to answer choice, yes/no and scoring questions without generating prose. It is an independent Jev-style approach; its L2 heads leave the base model’s weights unchanged. Use label-free L0 readouts or fit an L2 head for a particular question, with transformers or vLLM.

awesome-jev-by-typesafe collects sourced Jev use cases, prompts, patterns and starter code as an independent community resource. It explains Choice, Score and Noul for classification, routing and verification while keeping control flow in code. Explore Python and TypeScript examples; offline policy tests need no key, while live examples require TYPESAFE_API_KEY.

TypeLLM adds JSON Schema-constrained outputs to existing autoregressive models without changing their weights or architecture. Inspired by Jev, it retains free text and optional reasoning and supports dependencies between output fields. Serve a compatible model with SGLang, then submit context and typed questions through TypeLLMClient.

LocalJev translates typed decision requests into DiffusionGemma classification prompts and validates the generated probability output. Its Bun service exposes Jev’s response format through a standard OpenAI-compatible chat backend. Run it with Bun and an oMLX server, then direct TypeSafe SDK clients to the local API.

hippo-memory stores agent memories in SQLite, tracks provenance and lets feedback suppress incorrect memories. Its optional Jev reranker orders recalled memories and is disabled by default. Agents can remember and recall through the CLI or MCP, sharing memory across coding tools.

Rizzo Flow runs open decision models locally, returning choices, probabilities, rubric scores and numbers through a Jev-compatible HTTP interface. It reproduces Jev’s interface pattern, not its proprietary architecture or training; the project is independent. Serve a selected model with llama.cpp and point TypeSafe-compatible application code at the local endpoint.

Von scores independent questions in one forward pass for classification, yes/no checks and ordered ratings. It is a separate System One model with a TypeSafe-compatible API, running in-process or through HTTP. Use the Python or TypeScript SDK, or expose the local model with von serve.

DSCODE is a macOS coding agent whose terminal interface, CLI and scripts share a persistent session runtime. The repository includes a separate Jev client for OpenRouter’s alpha Decisions endpoint, supporting Choice, Score and Noul. Use sessions for editing, tests and agent handoffs; the Jev client requires an API key.

awesome-jev-tools catalogs public Jev projects and practices across routing, verification, ranking, agents and research. Inclusion requires a concrete Jev practice; generic projects that merely resemble decision models are excluded. Browse application categories, then inspect API calls, runnable checks, metric sources and licenses in linked projects.

Shapeshift turns typed text into event cards, checklists, timers, bill splits and other task-specific interfaces. Optional Jev classifies intent and signals; deterministic code handles dates and arithmetic. Offline keyword classification is the default. Type into the Bun-powered app, choose a card and save it; a server-side TypeSafe key enables Jev.

Jev Browser Use delegates navigation, clicks, toggles and scrolling while Codex handles text entry and final verification. Jev selects actions from observed controls using accessibility text; the existing Computer Use connection executes them. Requires Node.js 22+, a connected Codex browser and matching TypeSafe or OpenRouter credentials.

Distill is a terminal coding harness supporting Grok, Codex subscriptions and OpenAI-compatible models. Its Jev controls cover model tiers, skill suggestions and memory ranking; permissions remain the harness’s responsibility. Configure main, reasoning and utility models in Model tiers; the main model runs every session and step.

TipTour is a macOS menu-bar companion with Jev text tasks and a separate Gemini voice mode. Jev chooses locally detected controls; TipTour executes and validates clicks while screenshots stay local. Select a mode, supply its API key and grant permissions; building requires macOS 14.2+ and Xcode.

JevChat-Windows reads WeChat and KakaoTalk conversations with local OCR and offers candidate replies beside the chat window. Jev judges intent and tension and ranks three candidates; a separate language model drafts them. Configure a decision-provider key and a drafting-provider key on Windows 10 1903+ or Windows 11.

Jev Review scans Git diffs or complete codebases for correctness, security, reliability, compatibility and test coverage. Jev selects evidence, classifies mechanisms, scores severity and, when needed, uses Choice to select a reviewer. Code applies thresholds and workflow policy. Use Node.js 24+, Git and a TypeSafe key to generate JSON reports or inspect saved reviews locally.

Surf exposes Chrome navigation, page reading, clicks and screenshots as CLI commands for agents and scripts. Optional semantic commands ask Jev to find controls, rank page regions, verify outcomes and select bounded actions. Use deterministic commands for known flows; semantic.act explores toward a goal with confidence, freshness and budget checks.