Infrastructure
The services under the surfaces — routing, telemetry, scheduling, grounding, and identity plumbing.
The staff-facing control plane: fleet state, tenant administration, quota governance, and incident tooling. Internal only — not reachable with a customer account.
Alias resolution and failover for every inference request. Turns a stable model name into a healthy pool, on-prem first and cloud second.
The public edge: TLS termination, hostname routing, rate limiting, and security headers in front of every surface and the gateway.
Metering and observability for the platform: tokens, latency, cost, and health, recorded per request and attributable to a key.
The platform's own crawl and index of the live web — the corpus that grounded answers and Discover are built from.
Retrieval and citation: decides when an answer needs sources, finds them, and binds each claim to something a reader can check.
Places inference work on GPUs: which model is loaded where, how many replicas, and which request goes to which worker.
Buys interruptible GPU capacity when the fleet needs to grow, and drains it gracefully when the market takes it back.
The single source of truth for what things cost: model rates, tier multipliers, plan allowances, and the routing economics behind them.
Directory sync for organizations: users and groups provisioned and deprovisioned from your identity provider, so access follows employment.
The unified view of people and organizations across the platform: one identity for a contact, whether they appear in Mail, Comms, Commerce, or ERP.
Turns model output into real files: documents, spreadsheets, slides, and PDFs, stored with a stable link and a retention policy.
The execution layer for agents: runs tool calls in isolated sandboxes with an audit trail, so an agent can act on the world without being trusted blindly.
Durable memory across surfaces and sessions: the facts worth remembering about you and your work, readable and deletable.