Studio

Search•Automate•Create

Inputs

Fabric Collections

Semantic layer & governed data

Documents & Files

PDF, DOCX, HTML, MD, TXT, XLSX

Warehouses & Lakes

Snowflake, BigQuery, S3

User Prompt

Questions, roles, variables

Knowledge

Parsers & Loaders

Document parsing, table extraction

Chunker & Segmenter

Semantic segmentation

Embeddings Generator

Vector embeddings

Metadata Enricher

Tags, ACL, categorization

Vector Index

Vector search & metadata

Semantic Bridge

To Fabric metrics

Orchestration

Prompt Manager

Templates, persona, context

Context Builder

Retrieval & reranking

Model Router

Model selection & fallback

Tool Router

Tool selection & auth

Workflow Engine

Multi-step flows

Session State

Variables & cache

Execution

Retrieval Engine

Policy-filtered access

Reasoning & Planner

Multi-step planning

Function Calling

CRM, ticketing, APIs

Document Pipelines

Summary, translation

Citation Builder

References & anchors

Output Renderer

Structured answers

Delivery

Assistants Runtime

Multi-tenant runtime

Publishing

Web, Slack, Teams, API

Webhooks & Triggers

Scheduled tasks

Adapter Registry

Model & tool adapters

Channels

Web App

Direct interface

Slack/Teams

Chat integration

Embed/API

Custom integration

Studio SDK

TypeScript/Python

Signals

Task delivery

Safety & Governance

• Policy Gate

• PII Redaction

• Approvals

• Audit Logs

Guard enforcement

Observability & Evals

• Traces & Metrics

• Evals Engine

• Feedback Store

• Tuning

Quality & learning

InputsKnowledgeOrchestrationExecutionDeliveryChannels

AI workbench architecture: Knowledge, Orchestration, Execution, and Delivery. Guard enforces policy at retrieval and generation. Observability and Evals close the loop for quality, cost, and learning.

❓

📋

📊

🎧

📰