Tool and skill selection with decision models

Research runtime that lets a bounded decision model like Jev operate over large tool, memory and context spaces via a virtual option space.

Confidence-aware skill router that uses Jev to pick agent skills from large catalogues, declining or asking for review on weak matches.

Agent harness with 100 mocked tools measuring steps to finish tasks when the LLM picks tools versus when Jev picks them.

Experiments testing Jev ideas against baselines, starting with Jev as a top-10 tool router for MCP agents on three benchmarks.

Archived Go harness finding Jev tool steering does not reliably cut agent tokens in Claude Code or Codex; context trimming helped.

Claude Code hook that asks Jev which skill fits each prompt, plus the week-long log that led its author to remove it.

Benchmark of BM25, embeddings, rerankers and Jev for picking the right MCP tool among 525, plus an experimental Jev search engine.

Research code evaluating Jev as an independent per-tool relevance judge for building task-level tool spaces on TGB, BFCL and GTA.

Minimal agent loop where Jev selects each action and approves tool calls, a chat model fills arguments, and MCP tools execute.

Coding agent that runs discriminative Monte Carlo Tree Search, with Gemini proposing actions and Jev Choice and Score guiding the search.

Reproducible experiments and reports on what Jev can and cannot do, from skill routing to GPQA and housing prices.

Standard-library MCP server giving Codex advisory Jev tools for role and resource selection, with local validation and 85 offline tests.

Natural-language CLI where Jev picks which installed 'reflex' programs a sentence calls, running them only when sure enough.

Per-turn router: one ~350 ms Jev call picks model tier, effort, tools and skill, with a hard deadline and regex fallback.

Local MCP server putting Jev in the coding loop: route next steps, pick prepared tool calls, review patches and verify claims.

Vercel AI SDK layer that uses Jev to auto-approve, ask a human or block each tool call, and to prune irrelevant tools before a turn.

Unofficial LlamaIndex reranker and router where Jev scores retrieved passages and selects query engines, with BEIR results.

Chooses the model tier, skills and MCP servers for each coding-agent session with two Jev requests, then launches Claude Code or Codex.

One CLI to command a flock of robots with an LLM per robot, plus an optional decision LLM such as Jev that picks among known skills.

Elixir client for System One decision models (Jev, Clef, OpenAI Decisions, OpenRouter, self-hosted) returning full distributions.

LangGraph dental-clinic agent where a local Jev-style router picks which tools to bind each turn, halving LLM token use.

ComfyUI custom nodes that use Jev to select candidates, judge conditions, score text, extract numbers and choose local Skills.

Zero-dependency agent harness that sends each loop fork (act?, which tool, risk, done?) to a decision model and keeps the LLM for writing.

Minimal demo that matches a user goal to the best MCP server or capability in a catalogue using Jev, in two stages for large catalogues.

Small demo that routes a task to a research, code, browser, support or writer agent with Jev and shows a visual swarm map.

TokenRhythm's NeoHorse repo: open 4B/9B agent models plus NeoHorse-Jev-4B, a prefill-only Choice/Noul/Score decision model.

Jev-style skill router for OmO: frozen Qwen2.5-0.5B plus a small head scores the skill catalog and trims the system prompt to top-K.

CLM-8B, a contrastively trained System One model that embeds states and actions separately, served behind a TypeSafe-compatible API.

Open 2B non-autoregressive decision model on Qwen3.5-2B for agent tool routing and DOM actions, with BF16, FP8 and INT8 weights.

Open infrastructure for training, evaluating and deploying System 1 decision models, with calibrated LoRA checkpoints on a 27B backbone.