Inference Models, Routing & EU Residency Note
Curated capability pools, adapter-based model routing, per-company keys, and an honest transparency note on ADR-005 inference data residency.
Non-technical dispatchers shouldn't have to guess whether Llama-3.3-70B, DeepSeek-R1, or another endpoint fits a task. Instead, SCA abstracts raw model selection behind company-approved capability routing while keeping full transparency for IT administrators.
Adapter-driven inference & capability pools
Under Settings -> AI Models, Company Admins and Platform Admins manage:
- Capability tiers: Fast operational Q&A, balanced synthesis, reasoning-heavy customs triage, and SearXNG-backed web enrichment.
- Health & latency visibility: Server-side inference adapters isolate vendor API semantics behind one clean contract, allowing companies to switch endpoints or bind company-specific encrypted credentials without touching specialist instructions.
Transparent ADR-005 disclosure on DigitalOcean Serverless Inference
We state infrastructure compliance facts plainly rather than burying nuances in fine print:
Important residency boundary (ADR-005): While application databases, vector indexes, uploaded SOP files, and conversation logs stay isolated under your tenant workspace configuration, inference requests routed through DigitalOcean Serverless Inference are governed by DigitalOcean's inference endpoint routing and do not guarantee strict EU-only physical execution guarantees for prompt/completion payloads by default.
For customers requiring contractual EU-only model execution, SCA's inference adapter layer supports plugging in verified EU-region private/VPC-hosted inference endpoints during pilot deployment without changing any application workflows or specialist revisions.