User repository on GitHub
One local endpoint. The right model for every turn — enforced in code, not prompts.
- Forks
- 4
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
Jevonian
An experimental local proxy for coding agents that sends jevonian/auto turns to Jev to pick a model and thinking level after deterministic compatibility and quota filters, while explicit routes skip Jev.
- Stars
- 27
- Language
- Go
- License
- AGPL-3.0
- Reading time
- 3 min
- Published by
- Community
- Author
- xinyao27
- Last verified
- Sep 2026
About this project
Keep the coding agent you already use. Jevonian sits between your agent and your providers, chooses a capable model for each turn, respects quota and cache economics, and records the decision locally. No manual model switching between planning, implementation, and smaller tasks. Get started · Connect an agent · See how routing works. Start in two commands (Node 22+): Open add a provider and a routing brain, then connect your agent with a Jevonian API key and jevonian/auto`. The quickstart walks through each step.
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
Highlights
- Only jevonian/auto consults Jev; the plan, execute, utility, and chat aliases, an explicit x-jevonian-phase header, and real model IDs bypass it.
- With no brain configured jevonian/auto errors instead of guessing, and if every configured brain is unreachable the request fails with 502.
- Jev evaluates a structured snapshot of intent, tool results, error counts, context headroom, and cache penalties, and returns route and depth in one call.
- Low-confidence turns are marked in the ledger and response headers rather than rerouted; fallback channels are tried only when the primary channel fails.
- Each turn records provider, actual tokens, cache reads, and estimated cost; credentials live in ~/.config/jevonian/credentials.json at mode 0600.
Quickstart
bash
npm install --global jevonian
# then point your agent at http://127.0.0.1:8787/v1Watch out
AGPL-3.0-only. Needs Node.js and a provider key or sign-in; Jev receives recent messages and tool results (the optional fullPrompt setting can send the whole conversation), and cost and cache figures are estimates.
Questions about this entry
Short answers drawn from the linked project or page.
No, and the README says so directly: it is a local model router and protocol gateway, not an agent, not a task manager, and not local inference. Your agent owns the workflow and your provider runs the model. Jevonian routes requests, forwards them, and records what happened.
Not on its own. Cursor runs in the cloud and only accepts a public HTTPS URL for a custom OpenAI endpoint, so 127.0.0.1:8787 will not work. The README documents an explicit opt-in tunnel that publishes nothing but the /v1 path, using either a Cloudflare quick tunnel or a reserved ngrok domain.
A compact structured state rather than the transcript: user intent, tool results, consecutive error counts, context headroom, candidate capabilities, and cache switch penalties. One call answers both which route to use and how deeply to think. An optional fullPrompt setting widens it to the whole conversation.
Three documented paths. With no brain configured, jevonian/auto errors instead of guessing; the README describes heuristic phase classification as the fallback when every configured brain stays unreachable after retries; and low minConfidence marks the turn in the ledger and response headers without rerouting, since fallback channels are tried on failure, not on low confidence. No routing accuracy comparison against hand-picked models is published.
npm install --global jevonian then run jevonian; it serves at 127.0.0.1:8787, and on macOS it installs a LaunchAgent so the proxy keeps running after the terminal closes. Node 22 or newer is required.
Provider API keys and sign-ins are kept in ~/.config/jevonian/credentials.json with 0600 permissions, while generated Jevonian API keys are shown once and stored hashed.
Send an explicit route alias like jevonian/plan, execute, utility or chat, set the x-jevonian-phase header, or use a real model ID, which is pinned and never routed; routing.mode off makes the proxy pure pass-through.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: A usage router for Grok Bot
X: A prompt box that fills itself in
X: How Jev makes agents faster and cheaper
YouTube: CJ from Syntax demos browser use, classification, code review, a model router, and a chat bot with no LLM.
More Jev resources like this
A local gateway for coding agents that asks Jev which tool to call each turn and steers the LLM when Jev is confident, while every other request goes to your usual provider untouched.
Jev Tools & CLIs#community#typescript#gateway
Community
vinilana/jev-gatewayTypeScript
vinilana/jev-gatewayTypeScriptA TypeScript layer beside an LLM: JevSift selects approaches, JevLoop audits progress with typed actions, and JevLong watches sessions for stalls without steering them.
Jev Tools & CLIs#community#typescript#coding-agent
Community
Alex314618-create/JevRevTypeScript
Alex314618-create/JevRevTypeScriptA router that turns models, subagents, skills, MCP tools, CLIs, and plugins into one candidate set Jev answers with a single typed Choice, while the router code owns availability, permissions, and confirmation.
Jev Tools & CLIs#community#typescript#routing
Community
BillionsBobby/JevRouterTypeScript
BillionsBobby/JevRouterTypeScriptA local-first Herdr CLI that filters your agent subscriptions with fixed eligibility rules, then asks TypeSafe Jev to rank the remaining Cursor, Claude Code, Codex, or OpenCode options and pick a reasoning effort.
Jev Tools & CLIs#community#typescript#cli
Community
nidhi-singh02/agent-routerTypeScript
nidhi-singh02/agent-routerTypeScriptA Pi coding-agent extension that routes tools and skills with Jev: one pass activates only what a prompt needs, with opt-in model selection, tool-call guards, and compaction.
Jev Tools & CLIs#community#typescript#coding-agent
Community
TheoOliveira/pi-jevTypeScript
TheoOliveira/pi-jevTypeScriptA local proxy between Claude Code and the Anthropic API that picks the model and effort level per message with Jev, routing subagents whenever cheap but gating main-chat switches on prompt-cache arithmetic.
Jev Tools & CLIs#community#typescript#routing
Community
adarshmishra07/jcm-routerTypeScript
adarshmishra07/jcm-routerTypeScriptFrom 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.