Cost tracking
Put a dollar cost on every AI agent step
Agents spend real money on every LLM call. Ballast attributes that spend to each step of each run — so you see what an agent costs, catch the expensive steps, and cap runaway bills before they happen.
Attribution
Every step, itemized
A run isn't one charge — it's a stack of model calls. Cost attribution shows you each one.
run r_9c21 · cost attribution
| step | model | tokens | cost |
|---|---|---|---|
| classify_ticket | claude-haiku-4-5 | 1,204 | $0.0031 |
| draft_reply | claude-sonnet-4-6 | 2,847 | $0.0261 |
| review_gate | — | — | $0.0000 |
| notify_team | — | — | $0.0000 |
| total | 4,051 | $0.0292 |
How it works
Cost tracking, built into the runtime
Cost per step, not per month
100% of steps are cost-attributed. Every node records the model, tokens, and price it used, so cost is a property of the run — not a mystery you reconcile at the end of the month.
One number across providers
Anthropic and OpenAI live behind one interface with unified cost tracking. A run that routes different nodes to different models still rolls up to a single, comparable cost.
Rollups that answer 'which agent spent this?'
Per-step costs aggregate up to the run and the workflow, so you can see exactly which agent, which run, and which step drove spend — and optimize the expensive ones.
Pass-through, or bring your own keys
On paid tiers, LLM token costs pass through at cost, itemized separately from the platform fee. On Pro and Enterprise you can bring your own keys — in which case nothing is billed for tokens at all.
FAQ
AI agent cost tracking, answered
- What is AI agent cost tracking?
- AI agent cost tracking attributes token spend to each step of each run — model, tokens, and price per node — and rolls it up per run and per workflow, so you can see exactly what an agent costs instead of reading a single lump-sum provider bill.
- Does Ballast track cost across different LLM providers?
- Yes. Anthropic and OpenAI sit behind one interface with unified cost tracking, and each node can pick its own model, so a run that mixes providers still produces one attributed cost.
- How do I stop an AI agent from running up a huge bill?
- Every tier has hard caps with cap-and-notify by default: runs beyond your limit stop until you raise the cap. Auto-overage is opt-in per workspace, with per-run pricing shown before you turn it on.
Know what every agent costs.
Start free — every step cost-attributed, with hard caps on by default. Related reading: how to track the cost of every AI agent step.