Article
CommunityJev Integrations & Adapters
What a Burger Combo Taught Me About Agents
A Microsoft developer post compares two lunch-ordering agents, one gated by six Choice questions to Jev and one plain GPT-6-astra, to argue that typed decisions belong in front of a language model while the agent loop stays in code.
- Format
- Article
- Reading time
- 3 min
- Published by
- Community
- Author
- kinfey (Microsoft)
- Last verified
- Oct 2026
Highlights
- Runs a Jev Harness beside a GitHub Copilot agent; both use the Copilot Python SDK with gpt-6-astra and the same McDonald's China MCP service.
- Six Choice questions go in one system_one request, and the compiled plan sets fulfillment, meal, size, drink, quantity, and quote_only.
- A Python gate checks a minimum confidence threshold, 0.75 by default, verifies required fields, and narrows the tool set before execution.
- The pickup path exposes five MCP tools, from query-nearby-stores to calculate-price, and keeps the request as location context.
- The post separates Jev from model routers: GPT-6-astra stays in place and model switching is not implemented.
Watch out
A Microsoft community post by kinfey about a hybrid prototype, not a product: the comparison table says model switching is not implemented, the agent loop still runs, and the 0.75 threshold is a demo setting. It warns confidence is not accuracy and a low-risk classification is no substitute for sandboxes, authorization, or tests.
Questions about this entry
Short answers drawn from the linked project or page.
A single page with two execution paths for one order: a Jev Harness that first produces six bounded Choice decisions, and a traditional GitHub Copilot agent that interprets the raw request; both call the same McDonald's China MCP service with gpt-6-astra.
Six Choice questions about meal, quantity, size, drink, fulfillment method, and how far the system may proceed, all sent in one system_one request rather than six conversational turns, and evaluated in parallel within the request.
It compares confidence against a configurable threshold that defaults to 0.75, checks for missing fulfillment, meal, and drink information, and narrows the exposed tool set; the post calls the threshold a demo setting, not a safety standard.
No. A model router picks which model or endpoint handles a request; Jev decides what the request means and which business branch applies. Both paths still run gpt-6-astra, and the post says model switching is not implemented.
It is not a code generator, should not be the calculator, and cannot grant permission to charge a card or deploy software. Schema conformance, semantic correctness, and authorization are different properties, and selecting from a menu is no guarantee of the intended meaning.
Five MCP tools: query-nearby-stores, then query-meals, query-meal-detail, query-store-coupons, and calculate-price. The post sketches dependencies rather than requiring every call to run in that exact order.
Because the location is open-ended text that the six typed fields did not cover: passing only those fields left the executor unable to find a restaurant, and the authors kept the original request as location context instead of asking Jev to invent an address.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Jev in front of the HeyGen MCP
X: Playwright CLI + Jev vs Playwright MCP
X: A local MCP server that cuts computer use from 23s to 0.8s. Cartwright built Cortex, a local MCP server where…
X: A side panel that drives any site's WebMCP tools
More Jev resources like this
An MCP gateway that routes tool selection through Jev: for each request, Jev answers one yes/no question per server and per tool, and the LLM only sees the tools that passed. The repo's benchmark covers 649 tools at 1/32 of a comparable routing cost.
Jev Integrations & Adapters#community#github#mcp
Community
ini8labs/jev-mcp-routerPython
ini8labs/jev-mcp-routerPythonPython 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.
Jev Integrations & Adapters#community#agents#framework
Community
pydantic/pydantic-aiPython
pydantic/pydantic-aiPythonMIT-licensed Python tool router for coding agents that keeps obvious calls on a fast path, asks Jev (default typesafe/jev-1.13 via OpenRouter) for genuinely ambiguous choices, and can run bounded MCTS when multi-step consequences matter. Ships a CLI, MCP server and hook integrations.
Jev Tools & CLIs#community#routing#mcp
A Python toolkit that hands browser sub-tasks to Jev instead of the coding agent: one fast_run call lets Jev drive clicks and typing step by step, and the agent gets back a typed outcome or a typed reason to stop.
Jev Integrations & Adapters#community#python#browser
Community
danielnc/jev-browsePython
danielnc/jev-browsePythonEpistemic 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
wojciechwiesner/jit-contextPythonLocal tool-routing classifier for coding agents: a gateway and MCP tool that ask Jev which tool to call next, keep your agent's own model and login, and log every decision so you can start by observing.
Jev Integrations & Adapters#community#catalog-import#logicrw
Community
felpsdev/jev-classifierTypeScript
felpsdev/jev-classifierTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.