eve
Vercel's filesystem-first agent framework, whose experimental evaluate and autoModel paths default to Jev via AI Gateway.
Overview
eve is Vercel's filesystem-first framework for durable AI agents. Core capabilities live in conventional locations, an instructions file, tools, skills, channels such as HTTP, Slack and Discord, and scheduled jobs, making projects easy to inspect and extend. A CLI initialises a project and starts an interactive terminal UI, with models chosen from Vercel's AI Gateway.
According to the list summaries, eve ships Jev (typesafe-ai/jev via AI Gateway) as the default evaluation model in its experimental evaluate path, and an experimental autoModel feature uses it to pick a language model from an allowlist. The README reviewed does not describe these features, so details come from the listings. Projects are created with npx eve init, which installs dependencies, initialises Git and starts an interactive terminal UI, and the package bundles its full documentation so coding agents can read it locally.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which allowed language model should handle this request (autoModel)? | — |
| Choice | Typed evaluation questions via the experimental evaluate path(inferred) | — |
How it works
- Initialise an eve agent project
- Enable experimental autoModel or evaluate
- Jev via AI Gateway picks a model or answers evaluations
- Agent runs with the selected model
Strengths
- Jev available as a framework default
- Filesystem-first, inspectable agent structure
Caveats
- Jev features are experimental and not described in the README
- Requires Vercel AI Gateway
Getting started
npx eve@latest init my-agent.At a glance
- Type
- Service
- Status
- beta
- Language
- TypeScript
- Runtime
- Node agent framework
- Has code
- yes
Use cases and integrations
How curated lists describe it
Agent frameworks: Vercel's eve engine ships Jev as the default evaluation model (`typesafe-ai/jev`) in its experimental evaluate path.
yibie/awesome-jevVercel's agent framework. Experimental `autoModel` defaults to Gateway `typesafe-ai/jev` to pick a language model from an allowlist.
AnotiaWang/awesome-decision-modelsVercel's open agent framework, which ships Jev as the default evaluation model in its experimental evaluate path.
hellogumbo/awesome-jev
Frequently asked questions
How does eve use Jev?
As the default model for experimental evaluate and autoModel features, per the listings.
Is it production-ready?
The Jev-related features are experimental.
How do I start?
npx eve@latest init my-agent.
Related projects
LangChain's Python framework, which includes an optional TypeSafeClassifier runnable plus model-routing and risky-tool middleware using Jev.
PluginproductionPythonLangChain.js, which includes an optional TypeSafeClassifier runnable and middleware for bounded routing and tool-call checks with Jev.
PluginproductionTypeScriptYielded Agent: durable Effect agent threads in TypeScript, listed for a TypeSafe decision provider for typed question sets.
ProjectbetaTypeScriptComfyUI custom node that asks Jev noul, choice or score questions inside image-generation workflows, e.g. to pick a workflow template.
ProjectprototypePythonCommunity .NET 10 SDK giving one API for choice, score and assessment over Microsoft-Decision-1, Ollama or native Jev.
PluginbetaC#LangChain blog guide on wiring Jev into an agent harness as a dangerous-action classifier and model router.
Articlediscussion