zooqueenandzeekay
57b3ade7be
101 citations named eleven HIPs that no longer exist. They were removed deliberately — 3a52264d gives a reason for each, either that shipping code contradicts the document or that the thing was never built — so the references wanted repointing at whatever survived, not the files back. Each survivor was established from content and confirmed against the candidate's own title and text, never guessed from the number. Object storage went to HIP-0405, whose CR carries the same repository, endpoint and port pair the deleted standard specified. Observability went to HIP-0132, which corrects the deleted claim by name: one telemetry plane per node, not a sidecar. Where nothing survived because nothing was built, the sentence was reworded to stand on its own rather than given an invented citation. Two citations turned out not to be about their stated subject at all. HIP-0033 cited "HIP-0032 (CI/CD Standard)" — a number that was never CI/CD; the subject is HIP-0036, which already appeared as the child of the same diagram node. HIP-0107 cited it as "the vfs.Backend interface contract this HIP standardizes on", but the deleted text contains no mention of vfs and HIP-0107 defines that interface itself. (cherry picked from commit 32989a78d640e61446e26151892f1d8c4a3dd4b3)
5.1 KiB
hip, title, author, type, category, status, created, updated, requires
| hip | title | author | type | category | status | created | updated | requires |
|---|---|---|---|---|---|---|---|---|
| 0124 | Bring-Your-Own Provider & AI — Unified Dashboard and Usage | Hanzo AI Team | Standards Track | Platform | Final | 2026-07-07 | 2026-07-08 | HIP-0004, HIP-0018, HIP-0121, HIP-0123 |
HIP-0124: Bring-Your-Own Provider & AI — Unified Dashboard and Usage
Abstract
This HIP specifies the customer-facing composition of the Hanzo cloud:
a customer brings their own cloud provider, brings their own AI
models and keys, and sees one dashboard with one metered usage
ledger at console.hanzo.ai. It defines no new machinery — each leg
is an existing HIP, referenced not duplicated:
- BYO provider — attach compute (HIP-0121), scale it elastically on the customer's own credentials (HIP-0123).
- BYO AI — connect any model provider and key through the ONE unified AI provider interface, the gateway (HIP-0004).
- One console — per-model and per-product metered usage, credit
drawdown, and balance-floor gating in
console.hanzo.ai(HIP-0018 billing; HIP-0118 for the SuperAdmin cross-tenant board).
Composed, this is the resell-ready OSS AI cloud: the customer supplies the two expensive inputs — compute and model access — and the platform supplies identity, scheduling, metering, and a single pane of glass.
Specification
Leg 1 — Bring your own provider
The org connects compute through the ONE fleet surface
(/v1/clusters, HIP-0121): a managed cluster, a pasted kubeconfig, a
GPU box, or their own AWS/Azure/GCP/DO/Hetzner/Nebius account with
credentials sealed per-org in KMS. Visor (HIP-0123) provisions and
autoscales on that fleet — for BYOC, in the customer's account, so
their cloud bills them for nodes and Hanzo meters only the management
plane (HIP-0121's 1% / $1 / validator-free tiers). Nothing in this
leg is new here; conformance means no second attach surface and no
second scaler ever appears in a product UI.
Leg 2 — Connect your AI
The gateway (HIP-0004) is the unified AI provider interface: OpenAI, Anthropic, Google, self-hosted engines, and every other backend are providers behind one API, one model-routing table, and one credential store. BYO AI means the org registers its own provider keys (or points at its own hosted models) and every downstream product — chat, agents, ML pipelines — consumes them through the same gateway surface with the org's identity attached (HIP-0026). Platform keys and customer keys ride the same interface; whose key served a request is an attribute of the usage record, never a separate code path.
Leg 3 — One dashboard, one usage ledger
console.hanzo.ai is the single self-service pane (per-org); its
admin sibling is the SuperAdmin-gated cross-tenant board (HIP-0118).
Verified live this cycle:
- AI usage metering is real: per-model usage records (501 records on the reference org at verification), rendered with credit drawdown against the org's balance.
- Per-product usage is a server-side axis
(
?product=/?groupBy=product,hanzoai/cloud#159) — AI, compute, storage, and every metered product report into one ledger, one drawdown. - Billing is gated, not advisory: the balance floor returns HTTP 402 at the platform edge (HIP-0106 realized state; HIP-0018 semantics). Usage views and enforcement read the same ledger.
The conformance rule is the DRY rule: one metering path
(commerce/metering, per HIP-0121), one ledger per org, one invoice —
whatever mix of BYO provider, BYO AI, and platform-native usage
produced the charges. A product that meters outside that path, or a
dashboard that aggregates from anywhere but that ledger, is
nonconformant.
Decided vs shipped
- Shipped: the gateway provider interface (HIP-0004, live as the
api.hanzo.aisurface), the fleet attach surface and nominal-fee metering (HIP-0121), console per-model usage with drawdown and the per-product axis, and 402 balance-floor gating. - Staged (owned by the underlying HIPs, tracked there): the daily BYOC 1% and monthly device billing loops (HIP-0121 roadmap), cross-provider per-tenant scale execution (HIP-0123), and the unified cross-tenant fleet+revenue admin board (HIP-0121 roadmap, HIP-0118 surface).
References
- HIP-0004 — LLM Gateway — Unified AI Provider Interface (BYO models/providers/keys)
- HIP-0018 — Payment Processing Standard (billing semantics; 402 gating)
- HIP-0026 — Identity & Access Management Standard (the org identity on every usage record)
- HIP-0106 — Cloud — Unified Hanzo Binary (white-label surfaces; the realized metering + balance-floor edge)
- HIP-0118 — SuperAdmin & Tenant Isolation Model (the cross-tenant board and who may see it)
- HIP-0121 — BYO Compute Fleet & Metered Billing (attach surface, sealed credentials, billing tiers, one metering path)
- HIP-0123 — Visor — Fleet & Fabric Autoscaling Across Any Provider (elasticity on the customer's provider)
hanzoai/cloud#159 — per-product usage axis (?product=/?groupBy=product)
Copyright
Copyright and related rights waived via CC0.