Skip to content
Docs
EN

Šis puslapis dar neišverstas, todėl rodomas angliškai.

Changelog

API version 2026-10-04 (sent as X-Oryvox-Api-Version on every response).

2026-10-04: router v1.1, cache engine, observability, billing

API version 2026-10-04 (sent as X-Oryvox-Api-Version on every response).

  • Router v1.1: workload-specific score weights (realtime, coding, batch, reasoning, content, support), live provider health, session stickiness (X-Oryvox-Session), metadata.priority, agent steps (oryvox.step), workflow budgets (oryvox.workflow_budget). See Models and routing and Agents.
  • Cache engine: request classification (live data, personal, tools), TTL by content, semantic modes conservative / balanced / aggressive (0.97 / 0.93 / 0.88), entity checks, version-aware keys (oryvox.versions), cache tags and POST /v1/cache/invalidate (scope cache). See Cache.
  • Observability: GET /v1/requests filters (model, status, workflow_id, trace_id), traces with spans, versions and overhead_ms, GET /v1/metrics, trace headers X-Oryvox-Trace-Id / X-Oryvox-Parent-Span, incidents and alert rules. New key scope requests. See Observability.
  • Feedback loop: POST /v1/requests/{id}/feedback (rating 1-5) feeds measured model quality; optimisation recommendations in the dashboard.
  • Billing: prepaid credit with reservations (402 insufficient_credits) or postpaid with a credit limit; 10% markup on ORYVOX keys, no percentage on BYOK; Core plans Free (10,000 requests a month), Pro €49, Business €199 (SSO, audit log, Zero retention); project budgets per day and month with budget_action block / downgrade / warn; monthly invoices; plan allowance (402 plan_limit). See Billing.
  • Idempotency: replays carry Idempotent-Replayed: true; 422 idempotency_mismatch; 409 conflict while running or when the result was not kept.
  • Responses: tools and function_call output items, input content parts, usage.input_tokens_details.cached_tokens, item_id on deltas.
  • Test mode: ory_test_ keys run the whole pipeline with simulated answers (provider: "sandbox", oryvox.test: true), JSON and tool-call simulation, and errors on demand (oryvox-test:rate_limited ...). See Getting started.
  • Webhooks: incident.opened, incident.resolved and test.ping events; payloads carry version.
  • Data: Redacted retention mode; trace sampling.
  • SDKs @oryvox/sdk 1.1.0 and oryvox 1.1.0: cache.invalidate, requests.feedback, metrics.retrieve, request-log filters, header options (session, trace, parent span, cache off / scope), idempotentReplayed / idempotent_replayed, InsufficientCreditsError.

2026-10-03: Core V1

First public release of ORYVOX Core.

  • OpenAI-compatible API at https://api.oryvox.ai/v1: /chat/completions, /embeddings, /models.
  • /responses: simplified Responses API with typed streaming events.
  • /requests and /requests/{id}: request log with cursor pagination and routing traces.
  • Routing profiles oryvox-auto, -fast, -balanced, -quality, -max, -reasoning, -code; per-request routing and max_cost.
  • Exact and semantic response caching; token optimisation (optimization).
  • Automatic provider and model fallbacks.
  • Credits, soft and hard budgets, Idempotency-Key, versioned pricing, savings reporting.
  • Data policies: retention (Standard, Minimal, Zero), PII and secrets handling, provider and region restrictions, IP allowlists, key scopes, rate limits.
  • Bring your own provider keys.
  • Signed webhooks for credits, budgets, invoices and security alerts.
  • Test mode with ory_test_ keys.
  • SDKs: @oryvox/sdk 1.0.0 (Node.js / TypeScript) and oryvox 1.0.0 (Python).
Ar šis puslapis buvo naudingas?