Use case · Cost-conscious teams and solo devs

Stop paying frontier prices for every agent turn

Coding agents send a lot of requests, and most of them are routine. AutoJev routes those to cheaper models, keeps strong models for the turns that need them, and shows you exactly where the estimated spend goes.

Lower cost preference picks the cheapest capable model
Cost baseline estimates savings per request
Usage and cost by agent, model and provider
Your keys, your providers — AutoJev adds no markup

Download AutoJev Star on GitHub

Usage

Requests

9,412

Input tokens

142.8M

Est. cost

$38.20

Cache hit rate

61%

Daily tokens · last 14 days

Where agent spend leaks

One model for everything

Agents are configured with a single default model. Renaming a variable and redesigning a module are billed at the same rate.

No per-agent view

Provider dashboards show totals per key, not which agent or which workflow is driving the bill.

Switching is too much work

Trying a cheaper model means editing configs in every agent, so it rarely happens.

Middlemen add margin

Hosted routers are convenient, but some add fees on top of provider prices or require buying credits.

How AutoJev keeps costs down

AutoJev puts a local router between your agents and your providers. You enter each model’s input, output and cache prices per 1M tokens. Agents can call one cheap model directly as <provider>/<model_id>, or use a route that decides which model handles each request: weighted load balancing across inexpensive models, or intelligent selection with the Lower cost preference. There, AutoJev estimates the quality tier a task needs (with help from Jev when it is configured), then deterministically picks the lowest estimated cost among candidates that meet the capability requirements at that tier. Models with unknown prices are used only as a fallback.

Cost-aware routing does not mean bouncing between models mid-task. Session affinity keeps a conversation on its model, and under Lower cost AutoJev keeps the session model unless switching offers a confirmed saving above 20%, including estimated cache reuse. Every request is logged locally with tokens and estimated cost, so you can check the result instead of trusting it.

Setup recipe: a cost-aware default route

1
Enter real prices
On Models, fill in input, output and cache prices per 1M tokens (USD) for every model, plus tier and capability flags. Estimated cost and the Lower cost preference depend on them.
2
Create a cost-first route
Create a route with intelligent selection, the Lower cost preference, and a mix of inexpensive and strong models. Tools, vision, reasoning and context window are hard constraints, so a cheap model without tool support is never picked for a tool-using request.
text
Route: autojev/cheap Strategy: Intelligent selection Preference: Lower cost Candidates: deepseek/deepseek-chat, openrouter/qwen/qwen3-coder-flash, openrouter/anthropic/claude-sonnet-4
3
Set a cost baseline
In routing settings, pick a cost baseline — usually the model you would otherwise use for everything. AutoJev estimates savings by comparing the baseline’s estimated cost with the selected model’s for the same request. It does not affect selection or billing.
4
Point your busiest agents at it
Connect the agents that generate the most routine traffic to `autojev/cheap`, and keep a `autojev/quality` route for reviews and hard refactors.
5
Review Usage weekly
Usage shows requests, input/output tokens, cache read/write, cache hit rate, estimated cost, success rate and average duration, with a daily token chart and breakdowns by agent, model and provider.

See spend by agent and model

Totals, a daily token chart and breakdowns — all computed locally from your request logs.

Usage

Requests

9,412

Input tokens

142.8M

Est. cost

$38.20

Cache hit rate

61%

Daily tokens · last 14 days

Cost levers in AutoJev

Lever	Effect
Lower cost preference	Lowest estimated cost among candidates that meet the task’s capability and tier requirements
Load balancing with weights	Send most traffic to inexpensive models by weight, within priority groups
Per-agent model selection	Heavy-traffic agents on cheap routes, critical agents on strong models
Prefer local models	Use Ollama when tiers are equivalent — no per-token provider charge
Cache price and hit rate	Cache-aware cost estimates and visibility into cache reuse
Cost baseline	Estimated savings vs. the model you would otherwise use

Estimates, not invoices

Costs in AutoJev are estimates based on the prices you configured and the usage the provider reports. If usage or a needed price is missing, the cost is shown as unknown. Your provider’s bill is the source of truth.

Frequently asked questions

Does AutoJev charge anything or add a markup?

No. AutoJev is free and open source. You use your own provider keys, requests go directly to the providers, and AutoJev adds no markup.

How does the Lower cost preference choose a model?

It filters candidates by required capabilities and context, estimates the tier the task needs (Jev can contribute to that estimate when configured), and then picks the lowest estimated cost using your configured prices. Unknown prices are fallback only. Within a session it only switches when the saving is above 20%.

What is the cost baseline?

A model you choose for comparison. For each request AutoJev compares the baseline’s estimated cost with the selected model’s. It only affects the savings estimate, not routing or billing.

Can I see cost per agent?

Yes. Usage breaks down requests, tokens and estimated cost by agent, model and provider, and Logs can be filtered by agent.

Will cheaper models break my agent’s tool calls?

Tools, vision, reasoning and context window are hard constraints in intelligent selection, so a model flagged without tool support is not chosen for a tool-using request. Set capability flags accurately.

Keep exploring

FeaturesSmart model routingExpose routes like autojev/fast to every agent. Load-balance models or let Jev pick by cost, quality, speed or balance.FeaturesUsage & request logsTokens, latency and estimated cost for every request, broken down by agent, model and provider. Metadata only, stored locally.Integrations

DeepSeekRun DeepSeek in Codex and Claude Code. AutoJev converts Responses and Messages to Chat Completions on localhost.CompareAutoJev vs OpenRouterA local router across all your providers versus a hosted API with hundreds of models. AutoJev uses OpenRouter as a provider.

Updated 2026-10-14

Give every agent the right model.

Install AutoJev, add the providers you already pay for, and connect your agents in one click. Everything stays on your machine.