frost
Go CLI model router that uses one Jev analysis to recommend a model, harness and effort level, then selects deterministically.
Overview
Frost, a Haplab project, is a model router. Give it a prompt, from a one-line request to a full markdown document, and it recommends a model, a harness or API, and an effort level where supported, with an explanation and alternatives in readable text or JSON for agents and scripts. It recommends what to run without launching the model.
Jev assesses the task in one analysis request on the normal path; the routing decision is then made locally and deterministically in Go. The default policy chooses the cheapest configured profile that meets a quality floor, with options to prefer quality or speed. You configure the models, harnesses, APIs and effort settings you actually have, can refresh model data from online sources, and can account for subscription limits through an optional CodexBar integration. Structured results expose analysis, constraints, evidence and uncertainty, and decisions can be recorded and replayed offline against a different policy.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Task assessment questions (e.g. complexity, task type) used to select a model profile(inferred) | — |
How it works
- Submit a prompt to frost route
- Jev analyses the task in one request
- Go policy picks the cheapest profile meeting the quality floor
- Output recommendation, alternatives and evidence as text or JSON
Strengths
- One Jev call per route; deterministic selection
- Bring your own models and subscriptions
- Record and replay decisions offline
Caveats
- Requires a TypeSafe key
- Configuration needed to reflect your model access
Getting started
brew install marcus/tap/frost, copy the example config to ~/.config/frost, set TYPESAFE_API_KEY, then frost config check and frost route "...".At a glance
- Type
- CLI
- Status
- beta
- Language
- Go
- Runtime
- CLI
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A flexible and configurable CLI model router using TypeSafe Jev.
hellogumbo/awesome-jev
Frequently asked questions
Does Frost run the model?
No, it only recommends what to run.
How many Jev calls per route?
One analysis request on the normal path.
Can I replay decisions?
Yes, recorded decisions can be replayed offline against another policy.
Related projects
- jev-router★ 593
Per-turn model routing for Claude Code and Codex: wraps the real CLI and lets Jev send simple turns to fast models and hard ones to strong.
ProxybetaJavaScript TypeScript/Bun launcher that asks Jev which model and reasoning effort each Codex, Claude Code or Pi turn should use.
ProjectbetaTypeScriptClaude Code and Codex plugin where Jev judges each turn for model routing, delegation, evidence search and compaction.
ProjectbetaTypeScriptClaude Code, Claude Desktop and Codex plugin where Jev reads a task and ranks live models and reasoning effort by quality, speed and cost.
PluginbetaTypeScriptLocal pass-through proxy for Claude Code and Codex where Jev picks a model tier per human turn and only moves up within a session.
ProxybetaTypeScriptLocal OpenAI/Anthropic-compatible proxy where one Jev call picks the model route and thinking level for each coding-agent turn.
ProxybetaGo