Enterprise AI
API Service
Compliance-First Distribution
Distribute Claude / GPT / Gemini to your enterprise clients securely and compliantly. WORM audit · Dollar hard cap · Error redaction · Vault approle · Portal isolation · Multi-key failover. Built for FinTech, healthcare, and listed-company B2B distribution.
5 HARD LINES · 100% CODE-TRACEABLE · WORM OBJECT LOCK 365 DAYS · VAULT APPROLE ZERO LONG-LIVED TOKEN
Distributing AI APIs:
Compliance Is the Real Problem
You want to wholesale Claude / GPT capabilities to enterprise clients. But every client demands audit logs, budget controls, and data residency — standard panels simply can't hold up to these requirements.
PAIN 01
Immutable Audit Trails
Compliance teams require call logs retained 180-365 days with physical deletion protection. Row-level soft-delete in a regular database won't cut it.
PAIN 02
Budget Exhaustion Attacks
One misbehaving client with a large concurrent burst can zero out your upstream USD balance. Post-hoc chargeback recovery rates are extremely low.
PAIN 03
Upstream Key Leaks
Upstream API keys mixed into 5xx logs — if clients see them, the liability is yours, whether regulatory or contractual.
PAIN 04
Compliance Certification Gaps
What SOC2 / ISO27001 / HIPAA auditors need to see simply doesn't exist in a standard proxy. Every audit becomes emergency remediation.
Not the prettiest panel —
the one that survives audits
We build and operate an enterprise AI API relay purpose-built for compliant B2B distribution. Every capability maps to a code commit and observable metric — no slide deck numbers.
365
Day WORM Retention
MinIO Object Lock compliance mode physically prevents any overwrite or deletion, exceeding most regulatory requirements.
5
Hard Red Lines
WORM audit · Dollar hard cap · Error redaction · Vault approle · Portal isolation. Each red line has CI-enforced tests.
49
% HMAC Speedup
VK HMAC single-pass verification is 49% faster than naive implementation, meeting SLO recording rule p99 on the hot path.
Nine Capabilities,
Every One Commit-Traceable
COMPLIANCE
WORM Audit (Object Lock)
All call events, key issuance/revocation, logins, and content interceptions land in MinIO Object Lock compliance mode — 365 days, physically undeletable.
BILLING
Dollar Hard Cap (DoW Defense)
GCRA multi-dimensional rate limiting + atomic pre-deduction by max_tokens × model price before each call, preventing Denial-of-Wallet attacks.
SECURITY
Error Response Redaction
5xx responses never expose upstream body. Unified {error:{code, message, trace_id}}. Upstream API keys won't flow back to clients even if log-leaked.
COMPATIBILITY
OpenAI ↔ Anthropic Translation
Clients use the OpenAI SDK unchanged; backend calls Anthropic's native endpoint. SSE streaming, tool_calls, extended thinking — fully bidirectional.
RELIABILITY
Multi-Key Pool + Circuit Breaker
Per-Tier (RPM/TPM) key sharding, auto-failover when provider 5xx rate exceeds 30%. Weighted least-connections — clients feel nothing.
OPERATIONS
Request Inspector
Ops surface: replay metadata (tenant/vk/tokens/cost/latency) for any request by trace_id. Original content never exposed. Locate any complaint in 30 seconds.
AUTH
SSO / OAuth Enterprise Login
Google / GitHub / OIDC (Azure AD / Okta / Auth0 / Keycloak). Hand-written OAuth2 + JWKS validation — zero new dependencies.
PRIVACY
PII Redaction + Tenant Policy
Presidio-style recognition of email / national ID / credit card / API key. YAML per-tenant policy: block / warn / redact — your choice.
EFFICIENCY
Thinking + Batch at Half Price
Claude 4.6 Opus/Sonnet 300k token output + native thinking trace passthrough + Messages Batches endpoint automatically billed at 50% off.
Three Steps to Live,
Zero Client-Side Changes
Add Your Upstream API Keys
Add Anthropic / OpenAI official API keys to the Provider Pool via the admin console. Keys are written to Vault transit-encrypted at rest — the application process never touches plaintext.
SETUPIssue Virtual Keys to Clients
Generate a virtual key under the tenant — shown once only. Set allowed models, IP allowlist, and monthly USD budget. Clients swap the base URL to SACTL and change nothing else.
DISTRIBUTEMonitor Usage, Billing, and Audit
Dashboard shows QPS / cost / latency percentiles. Inspector replays request metadata by trace_id. Audit streams 365 days of WORM events. Five pre-built Grafana dashboards included.
MONITORChoose by Scale,
Compliance Included in All Plans
All plans share the full set of red lines and audit capabilities. Differences are deployment model, SLA, and support response.
Self-Hosted Community
Open-source self-hosted for teams willing to run their own ops
Free
Recommended
Managed Enterprise
We host it — for enterprises that need compliance certs but not an SRE team
Contact for Pricing
Dedicated / On-Prem
Deployed in your network — for finance, healthcare, and government
Custom
All plans include the full red-line feature set. Managed Enterprise supports a 30-day free trial — no credit card required.
From Onboarded to
First Virtual Key in 30 Minutes
Try Managed Enterprise free for 30 days. We connect your existing Anthropic / OpenAI keys, then hand your first client their sk-xa-* — the audit and compliance clauses in the contract are already handled.
SELF-HOSTED COMMUNITY · MANAGED ENTERPRISE · DEDICATED / ON-PREM