Local & offline models — page 10

A Jev decision layer for Customer Success — 22 typed decisions per inbound message, 14 per account, 10 per outbound draft. Zero dependencies, 38 tests, runs offline.

An input that becomes what you mean: one text box that morphs into the right UI as you type. Powered by TypeSafe Jev, works offline.

BISTEC Architect — Tauri desktop app: local MiniCPM builds the brief, Jev (TypeSafe via OpenRouter) makes BISTEC-aligned technology decisions, humans approve ADRs

Local playground for decision models in the shape of TypeSafe's Jev, running the open openjev replication (Qwen3.5 as an NLI cross-encoder). CPU-only, no API key.

Unofficial local Jev Router adapter for the Codex desktop app on macOS

Local System One-compatible decision server for Laya backends

Single-file Bun proxy: TypeSafe Jev API → local Laya (or any Jev-like backend)

Local Vampire Survivors bridge and TypeSafe Jev agent with recorded gameplay, action probabilities, latency, tokens, and cost.

Local-first web app that pulls a GitHub repo's issues and README, then asks Jev (TypeSafe AI's decision model) to rate each issue by complexity, criticality, effort and relevance. Filter, dismiss, sort by weighted priority and export as plain text. Keys stay in memory; results stay in your browser.

Local MCP completion checks and repeat-loop detection for AI agents. Jev BYOK, deterministic policies, and a free offline demo.

Strands agents that make small local LLMs (Qwen3 0.6B, MiniCPM5 2B, Qwen3.5 4B) answer like Jev, a System 1 model: typed yes/no, choice and score answers with probabilities. Benchmarked against Jev on the jevals suite, via logprob readout and written-out JSON probabilities.

One C# IDecider, three engines: TypeSafe Jev (cloud), any local model in a constrained-decoding harness, and a fake for tests. 'System One' bounded decisions in .NET.

Local text classification for Mac and coding agents: yes/no, choice and score answers in milliseconds.

Ask TypeSafe's Jev a plain-English question, get a calibrated typed answer. CLI + local UI + public Cloudflare Worker + Claude Code skill.

Fifteen live AI games: TypeSafe Jev vs. Claude, GPT, OpenRouter and Ollama models

Let TypeSafe Jev play Brotato: local game bridge, autonomous menus, and recorded runs with probabilities, latency, tokens and cost.

Per-turn model AND reasoning-effort routing for Hermes Agent on Ollama:Cloud, decided by TypeSafe Jev (a System One decision model, not an LLM). Rewrites the outgoing provider request itself via llm_request middleware. Ollama:Cloud only.

MacOS organizer for images, PDFs and videos with TypeSafe classification and a searchable local dashboard.

Lightweight local supervision and governance for AI coding agents, using deterministic checkpoints and sparse Jev evaluations.

Local job-search app — fetches live remote listings, scores each against your resume with the Jev model, and drafts cover letters for the real matches. Flask, SQLite.

A local playground for watching TypeSafe AI’s Jev model make autonomous decisions in small games

Multi-agent customer support API where the LLMs write the text and a typed decision model makes the decisions. Jev (TypeSafe System One) handles routing, response evaluation and retry/escalation; FastAPI + Ollama. Includes a 60-ticket benchmark against LLM free-text and structured-JSON baselines.

Behavioural anti-cheat research bench for Minecraft (Paper): mining telemetry, typed LLM questions, and an offline benchmark against classic X-Ray heuristics. Shadow mode only, never bans.

Let Jev judge. Let your code decide. Python issue-routing sandbox with labeled replay, an offline demo, and Kev support.