Devin-AXIS/jev-dsh-decision
Jev DSH 决策引擎|面向 Agent Harness 的结构化决策插件。原生支持 DeepSeek Harness,通过 iPolloWork 支持 OpenCode、Codex Harness。
What it solves
It provides AI agents with a structured decision-making capability. Instead of relying on a chat model for planning, it uses a specialized decision engine to help agents objectively select the best tools, skills, or responsible agents for a task and evaluate the quality of outputs based on specific criteria.
How it works
The project acts as a plugin for various agent harnesses (such as DeepSeek Harness, iPolloWork, OpenCode, and Codex Harness). It connects to the TypeSafe AI "Jev" decision model via API. When an agent needs to make a choice or a judgment, it sends a structured request containing the task and available options to Jev, which returns a structured result including the selection, probability, and confidence level. It does not execute the actions itself but provides the decision data for the primary agent to act upon.
Who it’s for
Developers and users of AI agent frameworks (specifically DeepSeek Harness and iPolloWork) who want to improve the reliability and objectivity of their agent's tool selection and output evaluation.
Highlights
- Structured Outputs: Returns probabilities and confidence levels rather than conversational text.
- Multi-Harness Support: Compatible with DeepSeek Harness (native plugin) and iPolloWork (which extends to OpenCode and Codex Harness).
- Non-Intrusive: Does not take over the main agent's planning or execution; it only assists in decision-making.
- Batch Evaluation: Capable of evaluating multiple independent questions or criteria in a single request.
Related
- Project
Floe-Labs/floe-guardfloe‑guard is a Python/TypeScript library that tracks the real USD cost of every LLM, speech‑to‑text, text‑to‑speech, and other AI‑vendor call in‑process, and can hard‑stop a call before it would exceed a user‑defined dollar ceiling. It ships with adapters for OpenAI, Anthropic, Gemini, LangChain, CrewAI, LiteLLM, Vercel AI SDK, and voice platforms (Pipecat, LiveKit, Vapi, Retell). A bundled public‑price cost map provides offline pricing; you can override it with your own rate card. Budgets can be persisted across processes via SQLite, and an optional hosted Floe service adds a Coverage Score and 7‑day history. The core API is `BudgetGuard(limit_usd)`, with `check()` (pre‑call) and `record()` (post‑call) methods, plus atomic `reserve_tool`/`settle_tool` for paid tools.
- Project
harnessclaw/harnessclawA desktop application for managing, chatting with, and operating AI agents and their skills via a centralized interface.
- Project
pax-beehive/dsh-hub-cliA Node‑based CLI that lets you capture a DeepSeek Harness (runtime, ordered plugins, config patch, env‑var declarations) as an immutable, versioned *Preset Release* on the public DSH Plugin Hub. Others can fetch and apply the preset with a single command, getting exact plugin versions, load order, and runtime pinned. The tool supports planning, atomic staged installs, diff/upgrade/rollback, and integrates with AI agents via a safe plan‑apply workflow. All releases are content‑addressed (sha256), secrets stay local, and the CLI is MIT‑licensed.
- Project
sapientorius/NebulynkNebulynk is a self-hosted team collaboration platform that combines chat, video meetings, and AI-powered meeting summaries while keeping data and infrastructure under your control.