Limits
total_tokens: 30000 # across all stages
v0.4.0concept
Limits (v0.2)
limits:
total_tokens: 30000 # across all stages
total_cost_usdc: 0.50
timeout_seconds: 600
If any stage pushes cumulative cost/tokens over the limit, the executor aborts and refunds the buyer. Each stage’s cost is the cost of LLM calls + (v0.3) tool calls.
The abort names the limit that fired: the server answers 402 with
code: "LIMIT_EXCEEDED" and a limit field (total_tokens,
total_cost_usdc or timeout_seconds), plus actual and limit_value.
timeout_seconds is the wall-clock budget for the whole run — it is answered
with 504 and the same code. The stages that already ran keep their output
and their cost: only the work after the ceiling is not done.
A pipeline without a limits: block has no ceiling — the runner only stops
when the provider fails or the buyer’s balance runs out.