the reference
Everything, named precisely.
No narration in this chapter. Seven pillars, nine terms, four numbers. Read it like an index — because it is one.
seven capability pillars · 268 features · 20 areas
The index.
- 01
Conversational AI & Grounded Chat
Streaming, source-grounded assistants over your documents, meetings, email and the web — with reasoning transparency, multi-model routing, and an auto-router that picks the right agent.
- 02
Agents, Skills & Automation
No-code agent builders, a visual DAG workflow studio, reusable and executable skills, and 30+ prebuilt industry automations that produce finished Word, Excel, PDF and PowerPoint deliverables.
- 03
Knowledge, Data & Meeting Intelligence
Image-aware document indexing, governed Data Products with a natural-language Text2SQL semantic layer, reusable Knowledge Hubs, and meeting intelligence that tracks decisions and action items.
- 04
Governance, Security & Compliance
Strict multi-tenant isolation, dual-layer RBAC, an immutable tamper-evident audit trail, multi-stage PII and injection guardrails, envelope encryption, and gVisor-isolated code execution.
- 05
Observability, Cost & ROI Analytics
Executive dashboards for adoption, reliability and cost, with credit-metered billing, three-tier usage attribution, and hours-and-dollars-saved ROI tracking.
- 06
Channels & Integrations
Meet users in Microsoft Teams and WhatsApp, connect Microsoft 365 and Google Workspace, expose a REST + gRPC developer gateway, and extend agents via MCP tool servers.
- 07
Extensibility & Platform Foundation
Model-agnostic LLM routing, an extensible node and skill registry, MCP tools, a resilient priority job queue, and a microservices architecture deployable in cloud or on-prem.
the vocabulary
Nine words that mean exactly one thing here.
- Agent
- A configured AI worker — instructions, knowledge, tools and model — built by describing it in plain language and tested live before it is published.
- Skill
- A reusable, executable unit of behavior. Authored once, tested in a sandbox, then shared across every agent that needs it.
- Workflow
- A visual DAG of steps with triggers, branches, loops and approval gates. At a gate it freezes its full state and resumes exactly where it left off — even days later.
- Knowledge Hub
- A governed, reusable knowledge asset — team, project, workspace, personal or domain — referenced by pinned version so reuse never silently breaks.
- Data Product
- A governed, human-reviewed semantic layer over live databases and spreadsheets. Plain-English questions become safe, read-only SQL — validated before publish.
- Workspace
- A composable, governed environment binding agents, knowledge, tools and data sources — with chat hard-scoped server-side to what the workspace enables.
- Step Trace
- The live, human-readable record of a run: every step, tool call and file produced — resumable and abortable while it happens, kept after it ends.
- Approval Gate
- The point where work pauses for a person — single, sequential, parallel or quorum — with full saved state, all funneling into one unified inbox.
- MCP tool server
- Governed, read-only tool servers that extend agents over the Model Context Protocol — and let any MCP-compatible agent tap the platform in return.
counted, not rounded
14
platform services
7
LLM providers, model-agnostic
30
governance controls
AES-256-GCM
per-chat envelope encryption
The rest is a conversation.