The Core dashboard
Every ORYVOX Core section in the app, what it shows and what you can change there.
Everything for ORYVOX Core lives in one menu group, ORYVOX Core, in the app's left sidebar. Which items you see depends on your role.
Overview
Savings at a glance: what your requests cost with ORYVOX against the same requests on a single model, uncached and uncompressed. Below, where the savings come from (prompt caching, smart routing, input compression, response cache), the efficiency score, response time, cache hit rate, error rate and fallback rate.
No traffic yet? Start in two minutes: create a key in Projects and keys, then point your OpenAI client at ORYVOX.
Requests
Every request with its routing decision, cost and saving. Filter to errors only or answers from cache, and hide test requests. Click a row to see why that model was chosen, how the saving was worked out, the tokens and the time. A mark next to the model means the router picked it; without it, the model was named in the request.
Observability
What happened, where and why:
- Traffic: requests, availability, p50 / p95 latency, time to first token, ORYVOX's own overhead, errors, retries, fallbacks and timeouts.
- SLOs: the availability objective, the error budget left and its burn rate.
- Provider health: a live score from the last 15 minutes per provider.
- Spend: provider cost, the customer charge, net savings and cost per request, by project and budget.
- Incidents: anomalies against your own 7-day baseline, your alert rules and provider outages. A problem that continues is one incident, with one alert when it opens and one when it resolves.
- Alert rules and webhooks: thresholds for latency, error rate, daily spend, cache hit rate and fallback rate; signed webhooks for incidents, security alerts, budgets and billing events (retried up to 8 times).
Executions
How each multi-step AI task was carried out: the plan, the model per step, validation, escalation, cost and time. Replay a run exactly, or with today's configuration, and compare the answers. The Router and execution lab shows what ORYVOX would do with a task (requirements, plan, cost estimate and routing) without calling a model.
Models
Every model with its real prices, context size and where it runs. The router chooses from its pool; any model can still be named directly. Quality starts as an estimate from the price tier until benchmarks and your real traffic measure it.
Cache
Answers reused, prompt tokens the providers cached, and what that saved, per cache: exact, semantic and the providers' prompt cache. See why requests missed and clear a project's cache. Stored answers are encrypted, kept per project and expire by content type.
Optimization
What your real traffic says could be cheaper, faster or better, with the evidence. Nothing changes until someone decides. A new router policy goes candidate → replay on past decisions → shadow on live traffic → canary → production, each step needing the evidence of the one before, and production can always be rolled back.
Playground
Compare ORYVOX Auto with any models on the same prompt: answer, cost and speed side by side. View code gives you the request to copy.
Projects and keys
A project holds its keys, routing mode, token optimisation level, caches, context and output caps, and budgets.
- API keys belong to one project and follow its policy. Limit what a key may call, allow only certain IPs, cap its daily spend, rotate it with a grace period or revoke it at once.
- Routing and policy: the default mode (Economy saves most, Max always uses the best models), token optimisation (Conservative, Balanced, Aggressive), tool pruning for agents with many tools.
- Caches: response cache, semantic match threshold, cache lifetime, prompt cache.
- Budgets: per day or month, block, downgrade to economy near the limit, or only warn.
Credit and usage
What you have, what you used, where it went and what the month will cost: credit balance, unbilled usage, a forecast, the cost without ORYVOX, and spend by project, model and key. Add credit, set a low-balance warning, change your Core plan, and see the ledger, where every credit and charge is kept in order and never changed.
Security
Security events (blocked secrets, masked personal data, unknown IPs, failed sign-ins, policy changes), company-wide access rules (IP allowlist, required two-factor, session length, SSO) and the audit log. See Security center.