Govern AI spend without storing prompts or responses.
CapHound gives you governance, auditability, and control using only the metadata a spend decision requires — never the content of a request.
Prompt-safe by design
We see the cost of a request. Never its contents.
Governing spend doesn't require reading prompts. CapHound records the metadata that makes a decision possible — and nothing a user or model actually wrote.
What CapHound records
providermodelteam · feature · customertoken countscalculated costpolicy decision & reasontimestamp
Everything a spend decision needs to be owned, explained, and audited.
What CapHound never touches
prompt textresponse textsystem promptsmessage contentembeddings inputanything a user typed
A design constraint, not a setting you toggle. There is no code path that writes prompt or response content.
The controls a review asks about
Built for the security team that has to sign off.
The controls a procurement review asks about — handled the way you'd expect of production infrastructure.
Data handling
CapHound runs on the metadata a spend decision requires — provider, model, team, feature, customer, cost, and the policy decision. Prompt and response content are never collected, stored, or logged. A design constraint, not a configuration option.
Access control
Role-based access across every workspace resource. Admins configure governance and manage access. Members view spend and respond to alerts within scope. Viewer access is read-only — for finance and executive stakeholders who need the numbers, not the controls.
Auditability
Every decision is written to an immutable audit trail — what was evaluated, which policy applied, what action was taken, who approved it, and when. Append-only by design: the complete history of AI spend decisions is reviewable at any time, and never rewritten.
Tenant isolation
Workspace-level data isolation, enforced independently of the application layer. Spend records, governance rules, audit trails, and connected workflows are scoped to the workspace and inaccessible to any other tenant.
Two-person approvals
Consequential actions — budget increases, seat revocations, exceptions — require a separate approver. A requester can never approve their own request, and every decision is recorded with who decided it and through which channel.
Enterprise security review
CapHound supports security review for enterprise procurement — documentation on data handling, access control, tenant isolation, and deployment, prepared for your security team. Available on request.
Deployment
Run it our way, or entirely inside your own cloud.
Two ways to deploy, one prompt-safe boundary. Choose the option that fits your data residency and compliance posture.
Run by us, on our infrastructure
The fastest path to value. CapHound is hosted and operated for you, with per-workspace tenant isolation and the same prompt-safe boundary. No infrastructure to stand up — only the metadata a spend decision requires ever reaches us, never request content.
- Operated and kept current by us
- Per-workspace data isolation
- Prompts and responses never collected
Runs inside your environment, for data residency
For teams with strict residency or compliance requirements, CapHound runs entirely inside your own cloud. Your spend data never leaves your environment — only aggregate spend figures are reported outbound for licensing. No inbound access into your network is required.
- Runs inside your VPC — your data stays put
- Outbound-only: aggregate spend totals, no content
- No inbound access into your network needed
Need security documentation?
We support security review for enterprise procurement — documentation on data handling, access control, the audit trail, and deployment, prepared for your security team.