01
One login, three modes
The interface reshapes into User Hub, Builder Studio and Command Center — end users, builders and admins on one platform.
The platform
One platform to answer, automate and act.
Connect your knowledge, data and applications once. Everything you build afterwards reuses them.
268
features across 20 product areas
30+
prebuilt industry automations
4-stage
AI safety guardrail pipeline
25+
PII entity types detected and masked
Three things it does
01
Answer
Grounded answers from your own documents, meetings, email and databases — never generic model knowledge.
- Retrieval grounded in your documents, projects, meetings and email
- Collapsible source citations on every answer
- Plain-English questions over governed business data — no SQL
- Access-scoped, so people only ever see what they should
02
Automate
A drag-and-drop workflow studio for multi-step processes, with branches, loops, retries and approval gates.
- No-code agent builders and a visual DAG canvas
- Triggers from email, forms, schedules and webhooks
- Approval gates that pause and resume with full saved state
- Reusable skills, tested live before they are published
03
Act
Real actions in your systems — update records, send communications, generate documents, trigger approvals.
- Create editable Word, Excel, PDF and PowerPoint deliverables
- Write back to ERP, CRM and internal systems
- Send email and channel messages on your behalf
- Every action written to an immutable audit trail
The foundation
Every automation makes the next one cheaper to build.
Knowledge, data, actions and governance are authored once and shared by everything above them.
Connected to your stack
It plugs into the systems the work already lives in.
- Models
- OpenAIAnthropic ClaudeGoogle GeminiAzure OpenAIDeepSeekPerplexityMoonshot / Kimi
- Channels
- Microsoft TeamsWhatsApp BusinessWeb chatInbound email (IMAP / Graph)
- Business apps & identity
- Microsoft 365 — OneDrive, SharePoint, OutlookGoogle Workspace — DriveSAP (Fiori automation)Nextcloud FilesMicrosoft Entra ID SSODow Jones Risk & Compliance
- Data & storage
- PostgreSQLWeaviate vector databaseMinIO / S3-compatible storageRedisNATS JetStreamCSV & Excel
- Developer
- REST + gRPC gatewayModel Context Protocol (MCP)Inbound & outbound webhooksPython & TypeScript SDKsLangfuse tracingPrometheus metrics
Why it holds up
Built for the parts that usually block AI.
02
Fail-closed data governance
Mandatory row filters are injected into every generated query at runtime and re-verified after rewriting. The model is never trusted with security.
03
Publish gates that can't be gamed
A data product cannot go live until its validation questions pass a live SQL safety check — and any config change forces re-validation.
04
Pinned-version reuse
Knowledge, agents and data sources are published as immutable versions and reused without silent breakage.
05
Credentials isolated from the AI
A separate service owns database credentials, runs strictly read-only queries, and returns credential-free results the AI layer never sees.
06
Human-in-the-loop everywhere
Approval gates in chats, workflows and agents pause and resume with full saved state, all funnelling into one inbox.