AE Agent
Token usage & cost
AE Agent shows how many tokens each conversation uses and an estimated cost, right in the bottom bar of the panel — so you always know roughly what a task is costing before the bill arrives.
The bottom bar
After each response, the bottom bar updates with the running totals for the conversation:
| ↑ Input | Tokens sent to the model — your messages, project context, and tool results. |
| ↓ Output | Tokens the model generated back — replies, tool calls, and code. |
| Estimated cost | A running estimate based on the provider's public per-token pricing. |
How cost is shown per source
What the cost field shows depends on how the model is connected:
| Source | Cost shown | Meaning |
|---|---|---|
| API key | Estimated $ | Pay-per-token, billed by the provider |
| Subscription | plan | No per-token cost — counts against your plan |
| Local | local | Free — runs on your own machine |
Why it's an estimate
Each provider counts tokens differently — tokenizers, context handling, and rounding vary between Anthropic, OpenAI, and Google. AE Agent computes its estimate from each provider's published pricing, so it's a close guide, not an invoice.
The displayed cost is an estimate based on public pricing. Your real, authoritative bill always comes from the provider (Anthropic, OpenAI, or Google) on your own account. Synthetic never charges for or sees your token usage.
Keeping costs down
- →Use a subscription or local model for everyday work to avoid per-token charges entirely.
- →Start a fresh conversation for unrelated tasks — long histories re-send context and grow the input count.
- →Use
Chatmode for quick questions; it typically uses fewer tokens than a full Agent run.