daf-jev
Composable Python toolkit for Jev: question builders, confidence gates, batch evaluator, calibration, CLI, MCP server and agent skill.
Overview
daf-jev is a modular Python client and decision toolkit for TypeSafe's System One API. It provides noul, choice and score builders grouped into a QuestionSet, sync and async clients with retries on rate limits and typed errors, and pure-logic composition patterns on top of the answers such as confidence gates. It also includes a concurrent batch evaluation harness, an MCP server, a CLI, an agent skill, and a reproducible manuscript and figure pipeline.
One wire contract supports several backends: hosted Jev plus self-hosted alternatives and OpenRouter decision candidates, through a provider registry. An explicit benchmark backend interface adds local priors and classifiers with prepared datasets, durable attempt receipts and separate quality and cost evidence. There is also a decision-point decider and a pipeline from Jev outputs into RxInfer.jl graphical models. It suits researchers and engineers who want more structure than a bare SDK.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this about a billing problem? | yes · no |
| Choice | What is the tone? | calm · frustrated |
| Score | How severe? | minor · noticeable · blocking |
How it works
- Build questions with noul()/choice()/score()
- Call JevClient.ask with a state string
- Apply confidence gates or composition patterns
- Batch-evaluate a corpus and calibrate thresholds
Strengths
- Retries, error mapping and usage accounting built in
- Multiple backends behind one contract
- Evaluation and calibration tooling
- MCP server and CLI
Caveats
- Large surface area for a client library
- Hosted backend requires a key
Getting started
uv sync, export JEV_API_KEY (or TYPESAFE_API_KEY), then use JevClient().ask(state, {...}).At a glance
- Type
- MCP server
- Status
- beta
- Language
- Python
- Runtime
- library, CLI, MCP server
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A Python toolkit for Jev requests, batch evaluation, calibration and MCP access.
logicrw/awesome-jev-projectsdaf-jev: composable Python toolkit for TypeSafe.
wh000wh000/awesome-jev-liveDaf-jev: composable Python toolkit for TypeSafe's Jev (System One) decision API — question builders, confidence gates, evaluator, calibration, CLI, MCP server, agent skill.
hellogumbo/awesome-jev
Frequently asked questions
How is it different from the official SDK?
It adds composition patterns, batch evaluation, calibration, multiple providers, a CLI and an MCP server.
Can it use local models?
Yes, its provider registry includes self-hosted backends alongside hosted Jev.
How do I authenticate?
Set JEV_API_KEY or TYPESAFE_API_KEY, or use a .env file.
Related projects
Rust CLI for Jev: ask yes/no, choice and rubric questions about text and branch on exit codes, JSON or MCP tools.
MCP serverbetaRustGo MCP server giving any client typed Jev decisions via OpenRouter or TypeSafe, with validated answers and no paid retries.
MCP serverproductionGoUnofficial hosted or local Jev MCP server giving agents a quick second opinion: it picks yes/no, scale or choice and returns probabilities.
MCP serverbetaTypeScriptRust CLI and library that asks Jev typed questions via OpenRouter and combines the probabilities with fuzzy-logic rules into decisions.
CLIbetaRustUnofficial standard-library Go client for Jev with generic typed Choice values and local confidence thresholds.
ProjectbetaGopip-installable Jev kit: a CLI for verify, screen, classify, extract, route and rerank, plus an MCP server for the primitives.
MCP serverbetaPython