Product
Everything between your app and the model.
Basient sits in the hot path of every LLM call your product makes — routing, governing, metering and auditing each request without adding meaningful latency or a single new SDK.
Unified API
One endpoint. Every provider.
Basient exposes a strict OpenAI-compatible surface — /v1/chat/completions, /v1/embeddings, streaming via SSE — and translates to OpenAI, Anthropic, Google, Azure, Bedrock, Vertex, Together, OpenRouter and more. Your clients never change; your providers can.
- ✓Drop-in: any OpenAI SDK works unmodified
- ✓True streaming: token-by-token SSE, backpressure-aware
- ✓Deterministic errors: one error envelope, provider details only in observability headers
Routing & failover
Milliseconds matter. So does the fallback plan.
Route by latency, cost, model quality tier or explicit policy. When a provider degrades, Basient fails over mid-conversation — and if a stream dies after first byte, the failure is terminal and visible: an error frame, a clean [DONE], an outcome=partial record. Never a silent retry that double-bills you.
- ✓Health-checked provider pool, per-region
- ✓Policy routing: pin models per tenant, per flow, per environment
- ✓Circuit breakers + hedged requests for p99 control
Multi-tenancy & governance
Isolation you can show the auditor.
Every query in Basient carries a tenant id by construction — the query layer makes cross-tenant access structurally impossible, and every route is covered by a leakage test in CI.
- ✓Per-tenant API keys, budgets and model allowlists
- ✓Budgets per (flow, client): hard caps and soft alerts
- ✓RBAC with least-privilege roles; SAML SSO + SCIM
- ✓Approval flows for config changes, full change history
Observability & cost
Every token accounted for. In integers.
One immutable usage record per request — coalesced at end of stream, idempotent end-to-end. Costs in integer micro-USD (no float drift), FX-dated conversion, budgets reconciled against a durable ledger.
- ✓Per-request traces (OpenTelemetry), latency breakdowns
- ✓Cost per tenant / flow / model / day, exportable
- ✓SIEM export: S3, webhook, or direct query
Deployment
Run it where your lawyers are happiest.
| Mode | Regions | For |
|---|---|---|
| SaaS | EU · US · BR | Fastest start; residency pinning per tenant. |
| Dedicated | Your region | Single-tenant cell, your keys. |
| BYOC / self-hosted | Your VPC | Data never leaves your perimeter. |