可観測性と運用

自前で運用するシステムで、Python 戦略、バックテスト、模擬取引、実取引、監視を扱います。 Jev は注文、戦略、エクスポージャー、保有ポジション、予算を評価し、型付きの選択結果と確率を返します。

AI アプリの追跡規約と計測機能を提供し、モデル呼び出し、検索、外部ツールを扱います。 TypeSafe 連携は systemOne を span として記録し、JSON 入出力、モデル、トークン使用量を含めます。例は Choice、Noul、Score に対応します。

DSCODE は macOS の開発エージェントで、端末 UI、CLI、スクリプトが永続セッションを共有します。 OpenRouter の alpha Decisions API 用 Jev クライアントを含み、Choice、Score、Noul に対応します。 セッション内で編集、テスト、他エージェントへの引き継ぎを行います。Jev クライアントには API キーが必要です。

TypeSafe Mario は、エミュレーターのテレメトリーと RAM 由来の観測を使う実験的な NES 操作ループです。 Python 3.13、TypeSafe キー、ローカルのゲーム環境を用意し、state-demo を確認してから play を実行します。

Local Context Compiler (lcc): clean, dedupe and compact prompt context before it reaches the model, then report every block dropped, the cache tokens a pass invalidates and when pruning pays off. Runs offline with a local 1K decision model. MIT, no API key, zero telemetry.

Confidence-gated incident routing with TypeSafe Jev

A unified, async Python framework for LLM applications—chat, autonomous agents, RAG, and sandboxed code execution. Supports OpenAI, Anthropic, Gemini, Ollama, and more through a single API. Features an 8-phase cognitive reasoning cycle, Docker-based isolation, human-in-the-loop approvals, session management, and comprehensive observability.

Deterministic-first cognitive orchestration with bounded TypeSafe JEV supervision, evidence-aware escalation, and privacy-minimizing telemetry.

An autonomous risk-monitoring MCP server. Define a risk profile — a topic, the geographic locations to watch, and the policy triggers that matter — and the server runs an agentic sweep: live web search scoped by a judgment model (Jev), source scoring, and a Markdown briefing you can read from any MCP client.

A universal, sub-100ms FOSS gaming bot framework powered by OpenCV for lightning-fast visual telemetry and Jev System 1 AI for non-autoregressive decision loops.

Single-agent Pi coding coprocessor with Jev semantic gates, baseline-to-current diff review, and append-only observability telemetry.

Autonomous PlayStation 2 AI Agent with real-time visual telemetry HUD powered by TypeSafe Jev System One

Three composable judgment pipelines on TypeSafe's Jev: support-ticket triage, observability alert triage, and a deploy-risk gate.

ForgeOS Browser - a private, local-first browser built for humans and prepared for trusted AI agents. Educational prototype. Zero telemetry.

一个用TypeSafe Jev打简化掼蛋的demo；TypeSafe Jev plays simplified Guandan — LLM decision-making over dynamic legal card actions with partial observability.

Jev Guard is a real-time Windows process monitoring application powered by TypeSafe's Jev AI model. It tracks newly launched processes, logs event metadata (paths, command lines, and parent processes) to a local SQLite database, and classifies risk levels—labeling events as benign, suspicious, or malicious on a live interactive dashboard.

Trace DeFi incident exposure through sourced portfolio dependencies with optional Jev classification.

Cheap fail-open semantic edge layer for Jev (TypeSafe System One): decision service + confidence policy + decision logging + blind eval harness + skill router over 1,000+ skills. Stdlib-only Python.

Experiments with Jev (TypeSafe classifier via OpenRouter): an incident-triage agent with Jev as tool and gate, and a smart model router compared against an LLM router

Local observability proxy and playground for TypeSafe Jev

Jev-powered operations decision support for Backstage: readiness checks, incident triage, change review, template and team recommendations, and catalog reranking.

Live-stream encoder settings from telemetry and history: TypeSafe's Jev decision model inside a deterministic safety envelope. Deno, Node, edge runtimes.

Conversational agent workspace with guided onboarding, voice, evaluations and monitoring.

AI Eval Platform — an evaluation and quality-engineering layer for LLM, RAG and agentic systems, with production observability integrations.