One endpoint for frontier, open-weight, and multimodal models across supported providers. Give each client their own keys, budgets, and provider policy; get a white-label usage report you can rebill. Bring your own provider keys — we never mark them up — or use the prepaid wallet at a rate published on the pricing page.
curl https://gateway.webik.ai/v1/chat/completions \
-H "Authorization: Bearer wk_live_..." \
-H "Content-Type: application/json" \
-d '{
"model": "auto/code",
"webik": {"profile": "balanced"},
"messages": [
{"role": "user", "content": "Refactor this function"}
]
}'reserved $0.0640 up front → settled to the actual bill; the difference is released
| client | monthly cap | spent | status |
|---|---|---|---|
| acme-corp | $500.00 | $131.58 | active |
| northwind | $250.00 | $212.40 | 85% used |
| initech | $120.00 | $120.00 | stopped |
Change one base URL, keep your tools. Use a WebikAI lane like auto/code, use the smart passthrough auto, or pin an exact model — every
response carries the receipt above, and every client's spend stops at its cap.
providers · region · data policy · capabilities
your key, or wallet cost reserved up front
price · speed · quality · reliability
errors fall back to the next ranked route
which model · which provider · what it cost
Nothing passes your filters? A clear error names the filter — rules are never silently relaxed.
OpenAI/Anthropic-compatible. Point Cursor, Cline, or any SDK at one base URL and reach frontier, open-weight, and multimodal models through supported routes such as OpenRouter, Z.ai, DeepInfra, and approved direct provider APIs — one dashboard and one set of controls. Managed-wallet usage is billed by WebikAI; BYOK remains billed by your provider.
Your rules filter first: allowed providers, US-only or EU-only, data policy, required capabilities. Then WebikAI ranks the eligible routes by the priority you chose — or, when you use `auto` / `auto/smart`, passes your constraints into OpenRouter Auto. Every response returns which model and provider served it, and why. No black box.
How routing works →Bring your own provider keys: encrypted at rest, decrypted only at request time, and never marked up. Or skip provider signups with the prepaid wallet: provider cost + a published markup for your plan (see pricing). When the balance hits zero, requests stop — nothing goes negative, ever.
The same model can behave differently at different providers. We run deterministic probe suites daily and track error rates and latency from real traffic, publish the scores with sample sizes, and rank regressing providers lower. It's beta, we label it beta, and you can always pin an exact provider yourself.
See who's leading now →The prepaid wallet is a hard stop at zero — spend is reserved conservatively before we call the provider and settled to the actual bill. Per-client, per-project, and per-key caps are enforced the same way (for BYOK traffic the pre-call estimate is best-effort). Kill-switch any key or client instantly.
Each client gets isolated keys, budgets, and provider policies. At month-end, download a white-label usage report at your own margin — the client-facing numbers you choose to forward, with our underlying costs kept separate.
OpenRouter has the biggest catalog and great aggregation — we route through it ourselves, and tell you when we do. Use OpenRouter directly if you're one developer who just wants models. LiteLLM is free and self-hosted — use it if you want to run your own gateway and don't need client billing. Portkey leads on enterprise guardrails. Use WebikAI when you run AI for clients: per-client isolation, budgets, provider policy, one dashboard across BYOK and managed, and reports you can rebill. IDE auto modes optimize inside one tool; WebikAI gives agencies the route receipt, budget ledger, and client report that make those choices auditable.
When a client needs dedicated, region-locked, or private capacity, we'll stand up a dedicated endpoint on the same account and controls.
Join the private-endpoint waitlist