What you pay for
JevAgent Usage and Pricing
JevAgent has no subscription or per-agent fee. Jev costs $0.042 per million input tokens, its output is free, and Vidbyte adds a 10% markup on each run's Jev total, billed to your Vidbyte API balance. Your agent's own model is billed at its provider's rate; by default those calls go through Vidbyte, which adds a 2% fee, and in direct mode they go straight to your provider with no Vidbyte fee. See how to read a whole run's total below.
Your agent's model and Jev are priced separately
Your agent's model is billed at that model's normal rates, plus a 2% Vidbyte fee in the default mode. Jev is billed only for input, at $0.042 per million tokens; its output is free, and there is no separate rate for cached tokens.
Your agent's own model does the work: it writes replies and uses tools. Jev, TypeSafe's decision model, makes the quick decisions behind preflight questions, tool selection, specialist routing, and continuation. Jev's price applies to jev-latest, jev-preview, jev-1.14, and jev-1.13.
- Input
- Jev$0.042 per million tokens
- Your agent's modelThat model's standard rate
- Output
- JevFree
- Your agent's modelThat model's standard rate
- Cached input
- JevNo cache tier
- Your agent's modelWhere the provider offers one
- Vidbyte fee
- Jev10% markup on each run's total
- Your agent's model2% in the default mode; none in direct mode
- Billed to
- JevYour Vidbyte API balance
- Your agent's modelDefault mode: your provider when its own key is set, with the 2% fee on your Vidbyte API balance; otherwise your Vidbyte API balance. Direct mode: your provider only
| Jev | Your agent's model | |
|---|---|---|
| Input | $0.042 per million tokens | That model's standard rate |
| Output | Free | That model's standard rate |
| Cached input | No cache tier | Where the provider offers one |
| Vidbyte fee | 10% markup on each run's total | 2% in the default mode; none in direct mode |
| Billed to | Your Vidbyte API balance | Default mode: your provider when its own key is set, with the 2% fee on your Vidbyte API balance; otherwise your Vidbyte API balance. Direct mode: your provider only |
One usage report covers the whole run
When a run returns, its response holds one usage report for the whole run: your main agent's work, every Jev call, helper steps, and a chosen specialist. The report also splits the same calls into your agent's model and Jev, so you can see what each part cost.
- Main agent work, including continuations
- Counted in the report underYour agent's model
- Up-front checks: preflight questions and the specialist choice, in one Jev call
- Counted in the report underJev
- Tool selection
- Counted in the report underJev
- Writing clarifying questions
- Counted in the report underYour agent's model
- Continuation helpers: planning, reviewing, and gathering evidence
- Counted in the report underYour agent's model
- Done checks
- Counted in the report underJev
- A chosen specialist
- Counted in the report underYour agent's model, at the specialist's own model rates
| Part of the run | Counted in the report under |
|---|---|
| Main agent work, including continuations | Your agent's model |
| Up-front checks: preflight questions and the specialist choice, in one Jev call | Jev |
| Tool selection | Jev |
| Writing clarifying questions | Your agent's model |
| Continuation helpers: planning, reviewing, and gathering evidence | Your agent's model |
| Done checks | Jev |
| A chosen specialist | Your agent's model, at the specialist's own model rates |
Read a run's total cost
The run's usage report already adds every part together. Read it after the run returns.
Total one JevAgent run
1reply = agent.run("Summarize the renewal options for this account.")
2usage = agent.response.usage # the whole run, set when the run returns
3
4# Every model call the run made: your agent's loop, every Jev call, helper steps, and a chosen specialist.
5# Costs use provider rates from the SDK price list; Vidbyte's fees show on your API balance, not here.
6print("total:", usage.total.cost_usd, "complete:", usage.total.cost_complete)
7
8# The same calls, split into your agent's model and Jev.
9print("your agent's model:", usage.generative.cost_usd)
10print("Jev:", usage.decision.cost_usd, "for", usage.decision.input_tokens, "input tokens")How Jev is billed
JevAgent sends every Jev call through Vidbyte with your Vidbyte API key and bills your Vidbyte API balance. You don't need a TypeSafe account or key.
What you need — A Vidbyte API key with the models:invoke permission (admin and full-access keys also work) and at least one cent in your API balance. Your agent's model calls through Vidbyte also need enough balance for each call's worst case: the provider's cost plus 2%, or only the 2% fee when your own provider key is passed through. A missing or rejected key, or an empty balance, stops runs that use preflight checks, specialist routing, or done checks with an error instead of skipping Jev; skill preloading marks its skills unavailable and continues, but in the default mode your agent's own model calls then stop with the same key error.
How Jev usage is billed — Usage is grouped into runs and priced at Jev's rate plus a 10% markup on each run's total, not on each call. Each JevAgent run is its own billing run. You are billed in whole cents as a run goes, plus any remaining fraction of a cent when it closes. A run closes when the JevAgent run ends, or after an hour without activity. Jev calls made outside a JevAgent run share one run per API key per day.
Only reported usage is billed — You pay only for the usage TypeSafe reports, so a failed call, or one with no reported usage, costs nothing.
Your agent's model — Billed separately, and by default also through Vidbyte with your Vidbyte API key. When your model provider's own API key is set, Vidbyte passes it through: the provider bills you directly and Vidbyte bills a 2% fee to your API balance. Without it, Vidbyte pays the provider and bills your API balance the provider's rate plus 2%. In direct mode, the calls go straight to your provider with your own key, and Vidbyte bills nothing for them.
Limits to keep in mind
Keep these current limits in mind when you read a run's cost.
Provider rates only — The SDK's total uses provider rates. Vidbyte's markups show on your API balance, not in the SDK's total.
Latest finish attempt details — Per-feature details on the response keep only the latest finish attempt. The run's total still counts every attempt.
Unpriced models stop the run — A model without a listed price stops a JevAgent run with a usage error instead of returning an incomplete total.