Decision-Native Agent Runtime
Research runtime where every agent step is a bounded typed decision, with a paged virtual option space over tools and memory.
Overview
Decision-Native Agent Runtime is an open research runtime for bounded, non-generative decision models. Its central abstraction is a Virtual Option Space: the logical space of tools, memories, context blocks and refinements can be larger than the options that fit in one decision request, so the runtime pages and refines that space across requests. Jev is the first live backend, but the runtime API is meant to be backend independent.
According to the list summaries, the design also covers tool integration, speculative parameter proposals, top-k proposals from an external helper model's logits with Jev-controlled fallback, decision-aware hierarchical memory and dependency-aware replanning, plus a natural-language conversation prototype. A core profile runs without a GPU, weights or API key as a deterministic protocol smoke test, an optional 0.8B helper proposes candidates, and the CLI requires explicit plan and approve steps before any side effect. Documentation is partly in Chinese.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which tool, memory or context option from the current page should be selected?(inferred) | — |
| Choice | Which proposed tool argument or refinement should be accepted?(inferred) | — |
How it works
- Runtime pages a large option space into decision-sized requests
- Optional helper model proposes candidates or arguments
- Jev selects among bounded options
- Side effects run only after explicit :approve
Strengths
- Scales a finite decision interface to large option spaces
- Auditable, approval-gated side effects
- Backend-independent design
Caveats
- Research prototype
- Autonomous natural-language shell is a separate future target
- Partly Chinese documentation
Getting started
pip install -e ., then python -m jev_agent.cli --workspace ./agent_workspace; add --live --key-stdin for Jev.At a glance
- Type
- Service
- Status
- research
- Language
- Python
- Runtime
- CLI
- Has code
- yes
Use cases and integrations
How curated lists describe it
Research design for a Jev-native agent system: tool integration, speculative parameter proposals, external helper logits Top-k proposals with Jev-controlled fallback ,decision-aware hierarchical memory, and dependency-aware replanning.Feature:Jev naturallanguage conversation prototype using external helper logits and dynamic Top-k token selection.
hellogumbo/awesome-jevResearch runtime for an agent whose every step is a typed decision rather than generated text, with a paged option space and a replaceable decision-model boundary. Chinese.
MrJev/awesome-jev
Frequently asked questions
What is the Virtual Option Space?
A logical option space larger than one request, paged and refined across decision calls.
Do I need Jev to try it?
No, the core profile runs without weights or keys as a protocol smoke test.
Can it act on its own?
The current CLI requires explicit plan and approve steps for side effects.
Related projects
One skill and MCP server for Claude Code, Codex and Hermes where Jev gates large-model turns, routes models and guards Git steps.
MCP serverbetaPythonPython harness for open-ended goals: AI-authored tools and checks, Jev choosing the next tool and arguments from fresh state.
ProjectbetaPythonPython agent runtime where Jev Choice decisions pick tools and arguments, and low-confidence picks go to an LLM for review.
ServicebetaPythonDecision middleware for AI agents that uses local Laya to pick tools and judge results, falling back to an LLM when unsure.
ProjectbetaPythonSelf-improving multi-platform agent harness with an optional Jev companion for typed judgments and per-session model routing.
ServicebetaTypeScriptOmO/senpi agent plugin where Jev or Span-01 rates which skills and next tools fit the task and spots loops or completion.
Agent skillbetaTypeScript