Describe the run
GPU type, count, memory and expected hours.
Agentic · ships today
Price a run against
the GPU market
Live GPU offers from your connected providers, cheapest first, next to Cairn's own estimate. Read-only.
Ships as compute/price-a-run
Before a training or inference run is submitted, an agent queries the GPU marketplaces you have connected for offers that fit, ranks them by real hourly cost, and places them next to Cairn's pre-flight estimate. It rents nothing.
The pipeline
Quote the market
The cheapest-gpu agent queries connected providers (RunPod, Vast.ai, Modal and others) for fitting offers.
Compare to the estimate
compute.estimate prices the run from your own settings; the two sit side by side.
Decide
Submit through the engine you choose, or change the plan. The quote is recorded.
What the record keeps
The run description, the offers returned with their hourly cost, Cairn's estimate, and the decision.
Price before spend
A run is priced against the market before any job is submitted.
Read-only
The agent queries and ranks; it never rents.
Same gate everywhere
cost.gate turns the estimate into a deterministic go or no-go for any pipeline.
Ships as compute/price-a-run on the compute operator. Provider coverage depends on the connected marketplaces.
See how failure can trigger a reviewed improvement cycle →Run it on your own infrastructure.
Apache 2.0 licensed, self-hosted. cairn pipeline run compute/price-a-run