AI Agents

Eight Agents. One Workflow.

Purpose-built AI agents that handle ingestion, transformation, analytics, dashboards, data quality, and end-to-end orchestration. Describe what you need, they do the rest.

Meet the Agents

Each agent specializes in one part of the data workflow.

Orchestrator
Concierge AI
End-to-end pipeline orchestration from a single prompt
Transformation
SQL Copilot
Natural language to validated, safe SQL
Analytics
Analyst AI
Query and visualize your data with natural language
Ingestion
Data Source AI
Guided data source connection and schema inference
Generation
Pipeline Generator AI
Full pipeline from a single prompt
Dashboards
Dashboard Generation AI
Text-to-dashboard in one prompt
Editing
Dashboard Copilot AI
Iterative dashboard editing via conversation
Quality
Data Quality AI
Intelligent expectation generation and monitoring

Concierge AI

Describe what you need in natural language and the Concierge decomposes your request into phases, connects data sources, generates SQL transformations, builds the pipeline canvas, and deploys to Airflow. It handles the full workflow so you can go from idea to running pipeline in one conversation.

Example Prompt

"Connect my GCS bucket gs://sales-data, clean the CSV files, deduplicate on order_id, and create a monthly revenue dashboard."

Capabilities

  • Breaks complex requests into ordered steps automatically
  • Delegates to specialized agents (Data Source AI, SQL Copilot)
  • Visual pipeline canvas generation
  • Airflow DAG creation and scheduling
  • Follow-up conversations to refine results
  • Dashboard creation from analysis results

How It Works

  1. 1
    You describe your goal in plain English
  2. 2
    Concierge creates a plan with ordered phases
  3. 3
    Each phase executes: source connection, SQL generation, pipeline build
  4. 4
    You review and approve at each step
  5. 5
    The pipeline is saved, deployed, and optionally scheduled

SQL Copilot

Describe a transformation in plain English and the copilot generates SQL against your actual table schema. A 5-layer validation pipeline checks column references, type safety, and security before you see the result.

Example Prompt

"Remove duplicate orders by order_id, keep the latest updated_at, cast price to decimal, and filter out rows where status is cancelled."

Capabilities

  • Schema-aware SQL generation (knows your columns and types)
  • Column reference validation and fuzzy auto-correction
  • Type safety checks (TRIM on integers, COALESCE type mixing)
  • Security validation (blocks DROP, DELETE, ALTER)
  • Automatic error correction when SQL fails validation
  • Preview results before committing

How It Works

  1. 1
    You describe the transformation in the cleaning or aggregation node
  2. 2
    The copilot reads upstream schema (columns, types, sample data)
  3. 3
    LLM generates SQL with full schema context
  4. 4
    Validation pipeline checks references, types, and safety
  5. 5
    You see a preview of the output and can edit or re-prompt

Analyst AI

Ask questions about your data in plain English. The Analyst queries both Iceberg tables and BigQuery, generates charts, and can pin results directly to dashboards. It routes each query to the right engine automatically.

Example Prompt

"What were the top 10 products by revenue last month? Show me a bar chart."

Capabilities

  • Natural language queries across all data sources
  • Multi-source routing (Iceberg via DuckDB + BigQuery)
  • Automatic chart type selection based on results
  • Pin visualizations directly to dashboards
  • Follow-up queries to refine analysis

How It Works

  1. 1
    You ask a question (e.g. "What was revenue by region last quarter?")
  2. 2
    Analyst detects the right source (Iceberg table or BigQuery)
  3. 3
    SQL is generated and executed against your data
  4. 4
    Results are returned with an auto-selected visualization
  5. 5
    You can pin the chart to a dashboard or ask a follow-up

Data Source AI

The Data Source AI walks you through connecting a new source step-by-step. It browses buckets and folders, selects specific files, validates connections, infers schema, and hands off to the pipeline builder — all conversationally.

Example Prompt

"Connect to my GCS bucket gs://company-data and show me what files are in the transactions/ folder."

Capabilities

  • Interactive bucket and folder browsing
  • File-level selection within folders
  • Connection validation before proceeding
  • Automatic schema inference from file samples
  • OAuth token management for cloud sources
  • Handoff to pipeline builder with pre-configured source

How It Works

  1. 1
    You specify the source type (GCS, BigQuery, or REST API)
  2. 2
    The agent authenticates (OAuth for cloud, credentials for databases)
  3. 3
    It browses available resources (buckets, tables, endpoints)
  4. 4
    You select which data to ingest
  5. 5
    Schema is inferred and a data source record is created

Pipeline Generator AI

Describe a complete pipeline in one sentence. The generator gathers context from your connected sources, decomposes the work into ingestion, cleaning, and aggregation tasks, generates SQL for each node, lays out the canvas, and validates the result — all before you touch the builder.

Example Prompt

"Build a pipeline from my GCS sales CSV: deduplicate on order_id, calculate monthly revenue, and output a aggregation metrics table."

Capabilities

  • Automatic task decomposition from natural language
  • Source context gathering (schema, file formats, sample data)
  • SQL generation per node with validation
  • Visual canvas layout with connected nodes
  • Clarification flow when the request is ambiguous

How It Works

  1. 1
    You describe the pipeline you want in plain English
  2. 2
    The generator reads your connected sources and schemas
  3. 3
    It decomposes the request into ingestion, cleaning, and aggregation tasks
  4. 4
    SQL is generated, validated, and assigned to each node
  5. 5
    A complete canvas layout is returned ready to deploy

Dashboard Generation AI

Describe the dashboard you want and AI builds it — analyzing your tables, generating queries, picking chart types, and laying out widgets automatically. Ask a follow-up to refine, or approve and save.

Example Prompt

"Create a dashboard showing monthly revenue trends, top 10 customers by spend, and order count by status."

Capabilities

  • Understands what metrics and dimensions you need
  • Schema-aware query generation against your tables
  • Automatic chart type and layout selection
  • Handles simple and complex multi-widget dashboards
  • Clarification flow when intent is ambiguous

How It Works

  1. 1
    You describe the dashboard (e.g. "monthly revenue by region")
  2. 2
    AI identifies the metrics and dimensions you need
  3. 3
    Columns are resolved against your actual table schemas
  4. 4
    SQL queries and chart configurations are generated automatically
  5. 5
    A complete dashboard is returned ready to save

Dashboard Copilot AI

Open a dashboard and chat with the Copilot to refine it. Add, update, or remove widgets through natural language. Changes are validated and applied to the live dashboard instantly.

Example Prompt

"Change the revenue chart to a line chart, add a KPI tile for total orders this month, and remove the status breakdown."

Capabilities

  • Add, update, and remove dashboard widgets conversationally
  • Table discovery and schema-aware SQL generation
  • SQL validation before applying changes
  • Changes applied to the live dashboard instantly
  • Layout reordering and chart type changes

How It Works

  1. 1
    You open an existing dashboard and start a conversation
  2. 2
    The Copilot loads your dashboard state and available tables
  3. 3
    You request changes (e.g. "add a line chart for daily signups")
  4. 4
    Changes are validated and applied to the live dashboard
  5. 5
    You review the result and continue refining or save

Data Quality AI

Analyzes your table schemas and data patterns to generate data quality expectations automatically. Combines schema analysis, semantic pattern matching, and LLM inference with confidence scoring to surface the rules that matter most.

Example Prompt

"Generate data quality rules for my cleaning customers table."

Capabilities

  • Automatic schema analysis (types, nullability, uniqueness)
  • Semantic pattern detection (emails, dates, currencies)
  • LLM-powered business rule inference
  • Confidence scoring for every generated expectation
  • Approval routing: auto-approve high-confidence, review the rest

How It Works

  1. 1
    You select a table or pipeline layer to monitor
  2. 2
    Schema analyzer inspects column types and statistics
  3. 3
    Pattern detector identifies semantic patterns in your data
  4. 4
    LLM generates business-rule expectations from context
  5. 5
    Expectations are scored and routed for approval

Bring Your Own LLM

OptimaFlo is BYOLLM — you provide your own API key (OpenAI, Anthropic, Google, or any OpenAI-compatible endpoint). Your prompts and data never pass through our servers. The AI runs against your key, in your cloud.

Explore Further

See how the agents connect to the rest of the platform.

Pipeline Builder

See the canvas where Concierge deploys generated pipelines

BI & Dashboards

Where Analyst AI sends charts and visualizations

AI that does the work

Describe what you need. The agents handle the rest.

Try the AgentsContact Us

Eight specialized agents. Your LLM key. Your cloud.

OptimaFlo LogoOptimaFlo

AI-native data platform. From raw data to business dashboards powered by Apache open standards, visual pipeline building, and AI agents that handle the heavy lifting.

© 2026 OptimaFlo. All rights reserved.

We value your privacy

We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept All", you consent to our use of cookies. You can customize your preferences or learn more in our Cookie Policy and Privacy Policy.