How do Pydantic schemas become Jev questions in Pydantic AI?

An agent's output type is mapped onto Jev questions: the model docstring is the goal, each field docstring is that field's question, and enum member docstrings or BoolCriteria say what each answer means. Jev answers every field in one request, and each value stays inside its type with its own confidence.

What limits does TypeSafeModel enforce before calling Jev?

A pick-one question can carry at most 255 options - the route question counts every tool plus output types - and a rubric can have at most 10 levels. Going over is a 400 from the API, so the model raises a UserError instead, refusing the question before it is sent.

How do I install Pydantic AI, and what language is it in?

It is a Python SDK, installed from PyPI with `pip install pydantic-ai` or, if you use uv, `uv add pydantic-ai`. The Harness features come separately as `pydantic-ai-harness`. Documentation lives at pydantic.dev/docs/ai.

Can I swap the model without rewriting my agent?

Yes. The README describes every model as a string swap away: the model is given as a string such as `'openai:gpt-6-sol'` or `'anthropic:claude-fable-5-1'` when constructing the Agent, and the same agent code runs against another provider by changing that string.

What does attaching a durability backend change?

With a durability capability such as `TemporalDurability`, the same agent runs inside a durable-execution workflow: every model and tool call becomes a durable activity, so a run working through a background queue survives restarts, failures and long waits. DBOS and Prefect attach the same way.

How does an output type change what a run returns?

Give the agent an output type and every run comes back validated and typed, so your IDE, type checker and the LLM agree on the returned shape. In the README's example a `Sentiment` model with a `Literal` label and a bounded score comes back as `label='positive' score=0.9`.

Can model-written code be run safely?

The ecosystem includes Monty, described as a sandboxed Python interpreter that runs model-written code. Pydantic Logfire provides observability over the whole app, and instrumentation is plain OpenTelemetry, so any backend you already run works.

CommunityJev Integrations & Adapters

Pydantic AI

Python agent framework whose TypeSafeModel class runs Jev as a decision model inside Pydantic AI agents: Pydantic output-type fields become typed Jev questions answered together in one request, each answer carrying its own confidence, and tools can be offered as Choice options.

See highlights

Open github.com

Stars: 20.5k
Language: Python
License: MIT
Published by: Community
Last verified: Oct 2026

pydantic/pydantic-ai

Organization repository on GitHub

View on GitHub

How Python does AI. Agents, realtime voice, image generation, embeddings. Every model, every interface, typed end to end.

Forks: 2.9k
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

Agents, realtime voice, image generation, embeddings. Every model, every interface, typed end to end.

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

Tagscommunity agents framework structured-output

Works forJev for Classification Jev for Tool calling Jev for Decision gating

Used inJev for Software & DevTools

Languages: Python
Capabilities: Open sourceRequires API key

Highlights

Use TypeSafeModel by name: typesafe:jev-latest or a pinned version like jev-1.13.0.
Every output-type field is answered in one request, each with its own confidence.
provider_details['confidence'] returns the per-field probabilities from the response.
Caps: 255 options per pick-one question and 10 levels per rubric before a 400.
Install with pydantic-ai-slim[typesafe]; pip install and one environment variable get started.

Watch out

MIT licensed. Requires a TypeSafe API key, and the docs warn that jev-latest moves between releases, so a tuned confidence threshold should pin the version it was tuned against. Over-limit questions are refused with a UserError before the request is sent.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

AgentScope

33k

Alibaba's multi-agent platform, which implements native TypeSafe Jev classification through JevClassifierModel for binary, choice and score routing across agent pipelines. Python 3.11+, Apache-2.0, with a FastAPI agent service and web UI.

Jev Integrations & Adapters#community#multi-agent#framework

Community

agentscope-ai/agentscopePython

Eliza

19.6k

Open-source TypeScript agentic operating system: a core runtime, app for web and mobile, CLI, cloud services and first-party plugins. The directory's curated summary reports it integrates TypeSafe System One decision services for sub-100ms intent classification, action dispatching and confidence-gated tool execution.

Jev Integrations & Adapters#community#agents#framework

Community

elizaOS/elizaTypeScript

JarvisCore

192

Python multi-agent runtime that ships Jev natively in an optional extras install: agents ask typed Choice, Score and Noul questions through a shared decision client, the Kernel can pick specialist subagents by Choice, and RAG passages with a prompt-injection Noul above 0.70 are withheld.

Jev Integrations & Adapters#community#framework#mesh

Community

Prescott-Data/jarviscore-frameworkPython

JIT-JEV Context OS

Epistemic runtime for autonomous coding agents that uses Jev as a sub-100ms System 1 gate to shadow-evaluate capsule facts, propose evictions before bloat and prefetch memory between turns. Its fixture run reports 31.3% fewer turns and 52.6% fewer blind discovery operations, with a fail-open circuit breaker.

Jev Integrations & Adapters#community#agents#context-engineering

Community

wojciechwiesner/jit-contextPython

Laser

Focus mode for Omarchy and Hyprland that judges each screen against the task you declared and escalates from a red reticle to a window haze and, as a last resort, closing the drifting browser tab. Each check is one Jev on-task classification, with screenshots OCR'd locally and never uploaded.

Jev Integrations & Adapters#community#focus#linux

Community

iYassr/omarchy-laserPython

Qualm

macOS menu bar screen-time app that reads the screen as text through the Accessibility API and asks a System One model one Choice per user rule plus a Noul on whether the page is a payment, login or banking screen, stepping in only when a rule's probability clears its threshold.

Jev Integrations & Adapters#community#macos#digital-wellbeing

Community

RoderickQiu/qualmPython

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai