OpenJev (GPT-AGI)
Jev-compatible open System One engine with a Claude Code-style terminal REPL showing live probabilities, plus an experiment harness.
Overview
OpenJev (GPT-AGI) is an independent, Jev-compatible System One decision engine that serves typed Choice, Score and Noul decisions from open-weight models in one forward pass, with a probability for every answer and no JSON parsing. It reproduces the interface pattern of TypeSafe's Jev, not its undisclosed model or training.
Its distinguishing feature is a terminal REPL in the style of Claude Code where you can watch decisions happen, with probability bars, confidence and latency per decision; an openjev play maze mode replays real Jev decisions. It also ships a reference harness for experiments against the official Jev API, such as the pelican test (morphing a pelican into a bicycle sentence by sentence and watching P(bird) and P(vehicle) cross) and a decision landscape of 64 tickets where P(urgent) rises from 0.11 to 0.85. A roadmap describes phases toward the target UI.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Is the subject a bird or a vehicle? | bird · vehicle |
| Yes / no | Is this ticket urgent? | yes · no |
| Choice | Ticket priority(inferred) | P1 · P2 · P3 |
How it works
- Run the REPL or server with an open model.
- Ask typed questions about a state.
- Watch probability bars, confidence and latency.
- Re-run experiments against the official Jev API.
Strengths
- Visual REPL for understanding decisions.
- Reproducible experiments against real Jev.
- Jev-compatible interface.
Caveats
- Early phase; target UI not fully shipped.
- Does not reproduce Jev's model.
At a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- CLI REPL, HTTP server
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Jev-compatible System 开源Jev.
hellogumbo/awesome-jev
Frequently asked questions
What is special about this OpenJev?
A Claude Code-style terminal REPL that shows decisions, probabilities and latency live.
Does it run real Jev?
Its experiment harness can call the official Jev API; the engine itself uses open models.
Is it affiliated with TypeSafe?
No, it is an independent project.
Related projects
33-chapter Chinese tutorial on agentic RL and GRPO, including a chapter on when Jev-style external decision models beat training with RL.
ProjectresearchPythonBilingual research report on what Jev is, how its claims compare with independent evidence, and the open-source alternatives.
ProjectresearchDatawhale's Chinese Jev course: notebooks on the three primitives, translated recipes, evaluation and local Laya fine-tuning.
ArticleproductionPythonEvaluation suite for decision models with 23,900 pinned rows across nine task families and three primitives, plus a leaderboard.
ProjectbetaPythonLocal SvelteKit workbench and CLI for organizing, editing, running and comparing Jev experiments defined as JSON.
CLIbetaTypeScriptLlamaIndex notebook comparing Jev with open decoders, encoders and specialised tools on document-pipeline Choice tasks.
ProjectresearchPython