Start with the work you want to improve. We’ll shape the models, infrastructure, integrations and acceptance criteria around it.
A starting point for specialists, developers, labs and scoped departmental pilots.
A centrally operated platform for knowledge, applications and team workflows.
Who uses the system, what they need to achieve and how success will be measured.
Quality, context, latency and concurrency requirements tested against representative tasks.
Source selection, ownership, permissions, refresh expectations and retention requirements.
Service ownership, backup coverage, restore exercises and the agreed support process.
Private hosting and fully disconnected operation have different requirements. A disconnected deployment needs a verified plan for model packages, identity, certificates, software updates and each selected integration.
Some Office integrations currently depend on external assets. Air-gapped operation is qualified against the agreed feature set and environment before it is promised.
AI Workspace, Notebook and source-grounded answers, document workflows, App Studio, agents and schedules, and administration. Model quality, supported documents, workload limits and access controls are checked against your specific use cases.
Administrators prepare network sources for read-only snapshot imports. Automatic change detection and replacement indexing are not included in the current baseline. The PostgreSQL connector is a constrained pilot with query and data limits, rather than unrestricted database crawling.
Office agents are a pilot capability. Host versions, startup, sign-in, managed distribution and recovery need qualification for the chosen environment. Request a supported-host assessment before including them in a rollout.
Meeting integration and consent-based transcript processing are pilot work. Live recording and automatic transcription are not included in the current pilot. Model fine-tuning requires a separate model, compute and executor configuration; dataset preparation alone does not enable training.
Capacity depends on model size, context, simultaneous requests, document workloads and integrations. We agree and test the expected workload before specifying a system. Workstation and Server describe deployment options, not guaranteed throughput.
Review quality, access, performance, backup and recovery against the agreed acceptance criteria. Confirm operating ownership and the support scope before extending access or adding more use cases.
Tell us about your users, your knowledge and the work you want to improve.