Sovereign tier — routing profiles on your infra

Named routing profiles (Fast · Expert · Heavy · Code) decide the cheapest capable path for each turn. The chain runs on YOUR endpoints — local first, your own Heavy box next, your frontier key last. The edge returns only the decision; nothing but {route, profile, chain, params} ever leaves WAVE.

⚠ honest caveat: a local model is not a frontier model. Sovereign optimizes cost + data residency, not peak quality. Keep a frontier hop in every chain so hard turns still escalate — a local-only chain is allowed but flagged.

profiles

Fast local→frontier · temp 0.2 · 512 tok · 8s — search / summarize / direct

Code local→frontier · temp 0.2 · 1024 tok · 20s — code edits, low-temp determinism

Expert local→heavy→frontier · temp 0.5 · 1024 tok — your Heavy box absorbs mid-hard turns

Heavy local→frontier · temp 0.7 · 2048 tok · 45s — long reasoning, biggest budget

Defaults map per route; override per-call with route(prompt, profile="Expert") in any of the 5 SDKs, or per-license via /account/profiles. Machine-readable: /profiles.

zero-egress data flow

your agent → edge classify (embed only, optional via /vector) → decision → YOUR chain runs local→heavy→frontier on YOUR infra

The only thing that can leave your network is the prompt embedding for classification — and you can skip even that by sending a pre-computed vector.

EU AI Act & data residency

Because inference stays on infrastructure you control, Sovereign keeps prompts + outputs inside your jurisdiction by construction — useful for GDPR / EU-AI-Act data-residency obligations. WAVE sees a routing decision, not your data. (Not legal advice; your DPA + deployment region still govern.)

price

$0.0001/decision — same as pay-per-use. One decision per request regardless of how many chain hops you run; never a per-token bill on local hops (the edge never executed them). See pricing · local setup · agent context.