Product · v0.1
The instrumentation layer for AI spend.
Firedog sits between your workflows and your model providers. Every call is traced, priced, and attributed in real time — without changing how your engineers build.
Product surfaces
01
Run ledger
Every workflow run is itemised. Model spend, prompt weight, tool calls, retrieval cost, history overhead. Filter by team, workflow, customer, or model.
02
Component attribution
Costs don't roll up to a vendor bill. They roll up to the decision that caused them. Renegotiate from a position of evidence.
03
Drift & anomaly
When a workflow doubles in cost overnight, we tell you which node caused it and what changed.
04
Governance export
Auditable trails for Risk and Finance. CSV, API, or piped into your existing FinOps stack.
Pilot use cases
What funds and insurers
actually use this for.
Use case
Diligence at scale
Investment teams running document-grade LLM workflows across portfolios. We show which deal eats which budget.
Use case
Claims triage
Insurers automating first-pass adjudication. We expose the per-claim AI cost so it shows up cleanly in loss ratios.
Use case
Internal copilots
Compliance, research, and ops assistants. We tell Risk what each team's automated knowledge worker actually costs.
Questions worth asking
FAQ
- How long to integrate?
- Under a week for most stacks. We instrument at the orchestration layer, not inside each model call.
- Does it work with our model providers?
- OpenAI, Anthropic, Azure OpenAI, Bedrock, Vertex, Mistral, and self-hosted. If it sits behind an API, we trace it.
- Where does data live?
- EU-hosted by default. Self-hosted available for institutional customers. We never store the prompt or response payload — only the metadata needed to attribute cost.
- How is this different from AWS Cost Explorer or Apptio?
- Cloud bills tell you what you spent. They cannot tell you which prompt caused it. We work one layer deeper, where the decisions actually happen.
- Do we have to change our workflow code?
- Minimal. A thin SDK or proxy, whichever fits your stack. The team reviewing the deployment will not feel slowed down.
30-minute call · NDA available