AutoJev
Download

Feature · Usage & logs

LLM usage tracking and request logs for every agent

See which agent sent each request, which model answered it, how long it took and roughly what it cost. AutoJev records request metadata on your machine and turns it into usage totals — without storing prompts or responses.

  • Per-request model, route, tokens, timing and cost
  • Filter by date, agent, provider and status
  • Usage by agent, model or provider
  • Metadata only — prompts and keys are never logged
Download AutoJevStar on GitHub

Logs

TimeModelStatus
14:02:31deepseek/deepseek-chat200
14:02:18openrouter/claude-sonnet200
14:01:55ollama/qwen3:14b200
14:01:40openai/gpt-5-mini429
14:01:12deepseek/deepseek-chat200

Find out where your tokens actually go

When three agents share four providers, each provider dashboard shows only its own slice. You can't tell whether Codex or Claude Code burned through the budget, or which route quietly fell back to an expensive model.

Because every request passes through the AutoJev gateway, one place sees all of them. The Logs page shows each request as it happens; the Usage page rolls them up by day, agent, model and provider.

What each log row shows

ColumnWhat you see
Time / AgentWhen the request arrived and which agent or client sent it
Model / ProviderThe model that actually handled it and its provider
Route IDThe route the agent asked for, e.g. autojev/fast
Input / Output / CacheToken counts reported by the provider, including cache reads and writes
First byte / TotalTime to first byte and total duration
Est. costEstimated cost from the prices configured when the request was made
StatusSuccess or the error returned, plus whether it was streamed or JSON

The list refreshes every 5 seconds. Filter by date range, agent, provider and status, or search requests.

Open any request for the full story

Routing source and reason

See whether the model was chosen by Jev, local selection or load balancing — and the decision reason, including any fallback.

Every upstream attempt

When failover kicks in, details list each attempt with its provider, model, HTTP status and duration.

Token detail

Input, output, cache read and cache write tokens, plus output speed when the provider reports complete usage.

Label your own clients

Any BYOK client can use a placeholder key like autojev-local-<client>. AutoJev doesn't check it; the suffix names the client in logs and usage.

Usage at a glance

Totals for requests, input and output tokens, cache reads and writes, estimated cost, success rate and average duration, with a daily input/output token chart and a breakdown by agent, model or provider.

Usage

Requests

9,412

Input tokens

142.8M

Est. cost

$38.20

Cache hit rate

61%

Daily tokens · last 14 days

Tag requests from a custom script

Give each client its own placeholder key suffix and it shows up as a separate agent in Logs and Usage.

OpenAI-compatible env for a client named "nightly-eval"
OPENAI_BASE_URL=http://127.0.0.1:9527/v1 OPENAI_API_KEY=autojev-local-nightly-eval

Questions the numbers answer

Is the cheap route actually cheap? Filter Logs by route and compare the models it picked with their estimated cost. Did a provider start failing overnight? Filter by status and provider, then open a request to see every upstream attempt. Which agent uses the most cache? Switch Usage to "By agent" and compare cache reads and writes.

Because AutoJev knows both the route an agent asked for and the model that answered, you can audit routing decisions instead of trusting them blindly.

Private by design

  • Request logs store model metadata, status, timing, usage and estimated costs
  • Full prompts, responses and API keys are not persisted in request logs
  • Everything is stored locally in ~/.autojev/autojev.db; nothing is uploaded to an AutoJev cloud
  • Only requests routed through AutoJev are counted

Costs are estimates — on purpose

AutoJev multiplies the token usage your provider reports by the input, output and cache prices you configure for each model. If a provider doesn't report usage, or a price is missing, the cost is shown as unknown rather than guessed. Your provider's invoice remains the source of truth.

Frequently asked questions

How do I track token usage across multiple AI coding agents?

Route them all through AutoJev. Each request is logged with the agent that sent it, so the Usage page can break totals down by agent, model or provider in one view.

Does AutoJev store my prompts or code?

No. Request logs contain metadata — model, route, status, timing, token counts and estimated cost. Full prompts, responses and API keys are not persisted in request logs.

How accurate are the cost estimates?

They're as accurate as the token usage the provider returns and the prices you configure. Cost is computed with the prices in effect when the request was made; missing usage or prices show as unknown.

Can I see why a request went to a particular model?

Yes. The details view shows the routing source and decision reason, and lists every upstream attempt if the request failed over.

Where are logs stored?

In a local SQLite database at ~/.autojev/autojev.db on your machine. There is no AutoJev account or cloud sync.

Keep exploring

Use casesFor cutting agent costsRoute routine turns to cheaper models, set a cost baseline, and see estimated spend by agent and model. No markup.FeaturesDebug consoleTest models and routes in conversation or cURL mode and inspect the selected model, timing, tokens, cost and raw JSON.FeaturesFailover & load balancingWeighted pools, priority groups and sticky sessions, with automatic failover on 429s and 5xx errors and Retry-After cooldowns.FeaturesSmart model routingExpose routes like autojev/fast to every agent. Load-balance models or let Jev pick by cost, quality, speed or balance.

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.

Download AutoJevStar on GitHub

Free & open source · macOS · Windows · Linux