Is decide just a Jev call wrapper, or does the SDK model it differently?

Differently. decide is a third inference type declared per provider through an inferenceKinds field rather than inferred from behavior, so the SDK knows up front which providers can serve it. tryDecide returns null when no decision provider is configured, and returns null again on any failure rather than throwing.

Will a Jev outage or a bad key break my app?

No. Every internal caller fails open: model routing, context compaction, MCP tool selection, and RAG planning all fall back to their non-Jev behavior, and with no key configured the library behaves exactly as it did before. On Jev those internal decisions land in roughly 400 ms for about $0.00002 each, versus a full generation call.

Can I make decisions fully local instead of calling TypeSafe?

Yes. decide also serves the open-weights Laya model, Juspay's own XOR weights, and Perplexity's hosted Decisions API, and the first provider configured in that order is the one that runs. Ollama, LM Studio, and llama.cpp also work locally, with per-request credential overrides on top.

Why are the routing confidence thresholds asymmetric?

Because the two errors cost different amounts. Routing upgrades to a stronger model at 0.3 confidence but downgrades only at 0.6, on the reasoning that a wrong downgrade is worse than a wrong upgrade. The registry behind that routing covers 64 models and 132 aliases, with hundreds more reachable through aggregators.

Which decision providers can NeuroLink's decide use?

TypeSafe Jev, Laya, XOR, Perplexity's hosted Decisions API, and Cloudflare Clef, tried in that order with the first configured one running. Jev uses TYPESAFE_API_KEY and is also reachable through the Vercel AI Gateway.

How fast and cheap is a Jev decision in NeuroLink?

About 400ms and roughly $0.00002 per decision, according to the README, instead of a full generation call. Routing uses asymmetric thresholds, upgrading at 0.3 and downgrading at 0.6.

Which local runtimes can NeuroLink use?

Three fully local runtimes: Ollama, LM Studio, and llama.cpp, each with per-request credential overrides.

Jev directory

CommunityJev Repos & SDKs

NeuroLink

A Juspay TypeScript SDK that treats Jev as a third inference type beside generate and stream: providers declare which kinds they serve, and tryDecide returns typed boolean, choice, or score answers, null when no decision provider is set.

See highlights Open github.com

Stars: 157
Language: TypeScript
License: MIT
Reading time: 3 min
Published by: Community
Author: juspay
Last verified: Sep 2026

juspay/neurolink

Organization repository on GitHub

View on GitHub

The pipe layer of an AI nervous system — one interface connecting provider neurons to your application, across three inference types: generate, stream, and a calibrated decide (via TypeSafe Jev). MCP-native, voice (TTS/STT/realtime), RAG, memory, file processors. Powers Tara, Yama and Clairvoyance at Juspay.

Forks: 130
Last pushed: Updated Oct 2026

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

About this project

The pipe layer for the AI nervous system. AI intelligence flows as streams — tokens, tool calls, memory, voice, documents. NeuroLink is the vascular layer that carries these streams from where they are generated (LLM providers: the neurons) to where they are needed (connectors: the organs). → Docs · → Quick Start · → npm · → Blog.

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

Tagscommunity typescript sdk routing agents

Works forJev for Routing Jev for Tool calling Jev for Scoring Jev for Decision gating

Used inJev for Software & DevTools

Highlights

decide is declared per provider through an inferenceKinds field rather than inferred from behavior; of 40 unified providers, Jev alone serves decide.
Every internal Jev caller fails open — tryDecide returns null on any failure — so with no key configured the library behaves exactly as before.
Jev is used internally for model routing, context compaction, MCP tool selection, and RAG planning.
Routing uses asymmetric confidence thresholds, upgrading at 0.3 and downgrading at 0.6, because a wrong downgrade costs more.
A 64-model registry (7 providers, 132 aliases) backs metadata and context-window checks; Ollama, LM Studio, and llama.cpp run fully locally.

Quickstart

import { NeuroLink } from "@juspay/neurolink";
const pipe = new NeuroLink();
const decision = await pipe.tryDecide({
  state: { ticket: "Refund request, $42, first occurrence" },
  questions: {
    autoApprove: { type: "boolean", instructions: "Approve without human review." },
  },
});

Watch out

MIT-licensed and not affiliated with TypeSafe AI; needs at least one provider API key or a local runtime, and Jev decisions require a TypeSafe key.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.

FacebookClinicnotes

X: OpenRouter Decision Model Rankings

X: herdr + jev, an agent router

X: A group chat bot with decision room routing

From the Jev guides

Original write-ups on what this entry is for.

Decision gates in agent frameworks: what each integration ships

Oct 16, 2026

How Pydantic AI, Mastra, TanStack AI, DSPy, Spring AI, and LiteLLM expose Jev: where the typed decision sits in the loop, what each integration returns, and which big frameworks are only adjacent.

Jev Integrations & Adapters#agents#frameworks#integrations

Read article

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