What does jev-swap do?

It finds LLM calls in a TypeScript, JavaScript or Python codebase that are really decisions, converts them to TypeSafe Jev modules, and shadow-tests the conversion on recorded traffic so you can see the agreement before swapping production traffic.

What counts as a decision-type call?

Calls tied to an enum, a boolean, a small integer range or a yes/no prompt. The scanner understands OpenAI chat.completions and responses, Anthropic messages, the AI SDK with zod or JSON schema, LangChain structured output, and raw HTTP POSTs to OpenAI, Anthropic or OpenRouter.

Do I need an API key to try it?

No. npx jev-swap shadow samples.jsonl --mock runs the pipeline against a simulated Jev, clearly labelled in the report. The live proxy needs TYPESAFE_API_KEY unless you also pass --mock.

How does the shadow proxy decide when to apply a threshold?

Every request is forwarded to your real API unchanged, and matching requests are mirrored to Jev in parallel. Once a candidate has the minimum samples, 50 by default, and a threshold meets your target, the Apply threshold action writes it into the generated module.

What does it generate?

One Jev module per candidate with a decide() function and a confidence-gated fallback to the original LLM: .ts targeting @typesafe-ai/sdk 0.6.0 or .py targeting typesafe-sdk 0.7.1. Note that typesafe-ai on PyPI is an unofficial shim, per the README.

Streaming requests are forwarded but not mirrored, Jev's state is the last user message, and two candidates with identical schemas or prompts both match the first one scanned. Mirrored samples contain your users' inputs, so keep jev-swap-out/proxy-samples.jsonl out of version control.

What are the system requirements?

Node 22.12 or newer. Python scanning additionally needs Python 3.9 or newer on PATH as python3, overridable with JEV_SWAP_PYTHON, and it uses only the standard library.

Jev directory

CommunityJev Tools & CLIs

Jev Swap

An npm CLI that scans TypeScript, JavaScript and Python codebases for LLM calls that are really decisions, converts them into Jev modules with confidence-gated fallbacks, and shadow-tests them on recorded traffic before you swap.

See highlights Open github.com

Stars: 2
Language: HTML
License: MIT
Reading time: 2 min
Published by: Community
Author: 0xjba
Last verified: Oct 2026

0xjba/jev-swap

User repository on GitHub

View on GitHub

Find the LLM calls in your codebase that are really decisions, see what they'd save on TypeSafe Jev, and prove it on live traffic before you swap. TypeScript, JavaScript, Python.

Forks: 0
Last pushed: Updated Oct 2026

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

Tagscommunity catalog-import logicrw migration cli shadow-testing

Works forJev for Code Jev for Evaluation

Used inJev for Software & DevTools

Languages: TypeScript JavaScript Python
Platforms: CLI OpenRouter Vercel AI Gateway
Capabilities: Open sourceRequires API key

Highlights

Scan finds enum, boolean, small-integer-range and yes/no LLM calls across OpenAI, Anthropic, AI SDK, LangChain and raw HTTP client code.
Convert writes one Jev module per candidate as .ts or .py with a decide() function and a confidence-gated fallback to the original LLM.
Shadow replays recorded inputs and LLM outputs through Jev and writes REPORT.md; --mock runs the pipeline without any API key.
A live proxy at localhost:8787 mirrors traffic to Jev in parallel and shows agreement, latency, savings and a disagreement inbox.
Generated code targets @typesafe-ai/sdk 0.6.0 for TypeScript and typesafe-sdk 0.7.1 for Python, and requires Node 22.12+.

Quickstart

bash

npx jev-swap scan ./your-app
npx jev-swap convert
npx jev-swap shadow samples.jsonl --llm-price-in 1.25 --llm-price-out 10 --monthly-calls 1000000

Watch out

Grounded in the truncated repository README and GitHub metadata reviewed on 2026-10-15; the README's Public section was cut off. Proxy sample files contain users' inputs and should stay out of version control; streaming requests are forwarded but not mirrored, and Jev's state is only the last user message.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

Jev at home: voice pipelines, Assist routing, and sentinels

Oct 16, 2026

How Home Assistant conversation agents, macOS voice pipelines, and an advisory sentinel put a fast typed decision in the loop — one call per utterance, thresholds in code, and a daily token budget.

Jev Tools & CLIs#home-assistant#voice#automation

Read article

Search visibility, audited by decisions

Oct 16, 2026

Jev-based SEO tools split the crawl from the call: deterministic crawlers count and parse, while typed judgments rank fixes, place internal links, score search results, and track AI-answer mentions — with costs and caveats attached.

Jev Tools & CLIs#seo#search-visibility#audits

Read article

Jev on the Gateways: Vercel, Workers AI, and OpenRouter

Oct 12, 2026

The request shape barely changes across surfaces; the surrounding contract does. When to call the API directly and when a gateway, a Worker binding, or a router entry earns its extra hop.

Jev Tools & CLIs#gateway#vercel#cloudflare

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai

Jev Swap

Highlights

Quickstart

Questions about this entry

What does jev-swap do?

What counts as a decision-type call?

Do I need an API key to try it?

How does the shadow proxy decide when to apply a threshold?

What does it generate?

What are the limits?

What are the system requirements?

More Jev resources like this

From the Jev guides

From the Jev AI model community