Why keep thresholds out of the cache key in metajev?

A decision is treated as a pure function of state, question, provider and model; a threshold is a policy input, not part of the decision. Because metajev stores the full distribution under that key, moving an accept boundary across the whole recorded history costs only a pass over stored numbers, with no re-runs.

What can a metajev policy express?

An ordered list of rules, each with bands over confidence, a named label's probability, or an expected position on a score scale, narrowed by tag, type, provider or model selectors. Policies serialise to JSON with a fingerprint that ignores the name, and a Budget caps spending and call count before calls are made.

How do I install metajev today, given it is not on PyPI?

Clone the repository and run pip install -e ., or install straight from git with pip install "metajev @ git+https://github.com/YYTbit/metajev", which does the same without keeping the checkout around.

Can metajev use a local or offline model for decisions?

Yes. One interface has four backends chosen per question: TypesafeProvider for the hosted Jev endpoint, SGLangProvider, OpenAICompatProvider reading first-position logprobs, and MockProvider, an offline deterministic hash of the inputs for tests and dry runs.

How does metajev cap what a run is allowed to spend?

Attach a Budget carrying a usd_limit and a call_limit; it is checked before a call rather than after. A client built from several providers routes each question to the cheapest one whose recorded accuracy clears a floor, sharpening as outcomes accumulate.

What does the metajev MCP server expose to an agent?

metajev-mcp provides decide, record_outcome, replay, calibrate, boundary_table and store_summary, so an agent can read recorded distributions, replay a changed policy and inspect calibration without running the model again.

How does metajev keep an old decision defensible?

A store is a working database whose rows can be rewritten, so metajev adds a hash-chained ledger for decisions that must stand up after the fact. Decisions are immutable, so refreshing a changed model's answer means recording the outcome and re-asking.

CommunityJev Repos & SDKs

Metajev

Python library that records each Jev-class decision as the full distribution under a sha256 key of state, question, provider and model, then applies a separate policy. Thresholds, review bands, routing and budgets can change across the recorded history without new model calls.

See highlights

Open github.com

Stars: 1
Language: Python
License: MIT
Published by: Community
Last verified: Oct 2026

YYTbit/metajev

User repository on GitHub

View on GitHub

Cache the decision, compute the policy. Record typed decisions from Jev-class System One models under a key built from state, question, and model, then change thresholds, review bands, routing, and budgets without re-running anything.

Forks: 1
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Decision models answer typed questions with probability distributions. What you do about those distributions is a separate question, and most integrations answer both in the same function, with a threshold written next to the API call.

Quoted from the project's README. Read the full README.

Tagscommunity calibration policy caching

Works forJev for Decision gating Jev for Evaluation

Used inJev for Software & DevTools Jev for Customer Support

Languages: Python
Platforms: CLI MCP
Capabilities: Open sourceSelf-hosted

Highlights

Key = sha256(state, question, provider, model); thresholds stay out of the cache key.
Replaying a stricter policy over 500 recorded decisions costs no model calls.
Providers: TypeSafe, SGLang, OpenAI-compatible logprobs, and an offline deterministic mock.
Ships calibration reports, boundary sweeps, a hash-chained ledger and an MCP server.
metajev-mcp exposes decide, record_outcome, replay, calibrate, boundary_table and store_summary.

Watch out

MIT licensed and not yet on PyPI, so install it from the git URL. Storage grows with distinct questions rather than calls, and decisions are immutable, so the way to refresh a changed model's answer is to record the outcome and re-ask.

Questions about this entry

Short answers drawn from the linked project or page.

Related terms

Glossary definitions related to this entry.

More Jev resources like this

huncho

Dependency-free TypeScript SDK that turns Jev Noul, Choice and Score answers into named decisions: enter/exit thresholds with hysteresis, nested decision trees settled in one call, a JSONL journal, threshold replay without inference and Brier calibration.

Jev Repos & SDKs#community#typescript#decisions

Community

edgardcham/hunchoTypeScript

FastJev

Self-hosted Python SDK and System One-compatible API for runtime-defined Choice, Boolean and Score decisions on pinned open models, across Torch, vLLM, MLX, llama.cpp and WebGPU. Each result records probabilities, token usage, timing, model revision and prompt version, with committed row-level benchmarks.

Jev Repos & SDKs#community#sdk#self-hosted

Community

chengyongru/fastjevPython

Orcha Jev Judgment Spike

Open spike issue in the Orcha multi-agent orchestration platform: it proposes moving judgment out of expensive state-of-the-art models and onto Jev, which emits precise instructions for a cheaper executor to carry out, and requires findings, a cost model, and a recommendation as issue comments.

Jev Practices & Patterns#community#orchestration#cost-control

Community

open-orcha/orchaPython

S1 Ruby

Ruby gem that makes System One measurement a language primitive: psi prepares a state, judge/choose/score return calibrated distributions, and ? collapses them for control flow. Ships TypeSafe, protocol (/v1/systemone) and CUA providers behind one interface with its own spec suite.

Jev Repos & SDKs#community#ruby-gem#sdk

Community

innocentdiaz/s1_rubyRuby

Laya Typed-Decisions

The fine-tuned Laya checkpoint for four typed-decision workflows: 0.766 accuracy on 2,000 held-out decisions, above Jev 1.13.0's published 0.727 and the teacher self-agreement ceiling it trained past.

Jev Models & Reimplementations#community#open-models#local

Community

AutoGPT

187.5k

Autonomous agent platform with 187,504 stars for building and running agents from plain English or a visual builder. Its platform features first-class TypeSafe Jev decision blocks for typed routing, filtering, scoring and confidence-gated next-action dispatching.

Jev Integrations & Adapters#community#autonomous-agents#platform

Community

Significant-Gravitas/AutoGPTPython

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai