Data & AI Platform Connectors
Orca Runs Where Your Data Already Lives
Your warehouse is already the system of record and the system of governance. Orca executes inside it — no data movement, no shadow copies, no second control plane to secure. Agents inherit the policies your data team already wrote, and every action they take is written back to lineage.
Data & AI Platform Connectors
Snowflake
Governed agentic execution on the AI Data Cloud for retrieval, reasoning and write-back inside the account, under Horizon policy.
AI & Agents
Cortex AgentsCortex AnalystCortex SearchCortex AI Functions
Data & Compute
SnowparkOpenflow IngestionIceberg TablesStreams & TasksDynamic Tables
Governance
Horizon CatalogColumn-level LineageMasking & Row-access PoliciesAccess History
Databricks
Governed agentic execution on the lakehouse for agents, models and pipelines running under Unity Catalog permissions and lineage.
AI & Agents
Agent BricksMosaic AI Agent FrameworkAgent EvaluationModel ServingAI Search
Data & Compute
Delta LakeDatabricks SQLLakeflow PipelinesDocument IntelligenceMLflow
Governance
Unity CatalogUnity GatewayEnd-to-end LineageModel & Agent Access Control
What the connector actually does
Executes in-platform
Orca pushes work down to Snowpark and the lakehouse rather than extracting it. Data stays in your account, in your region, under your contract.
Inherits your policy
Masking, row-access and RBAC defined in Horizon Catalog or Unity Catalog are enforced on every agent call. No parallel permission model to maintain.
Proves what happened
Every prompt, tool call, query and write-back is logged with identity, cost and outcome — traceable back through lineage to the source column.
Warehouse-Native AI, Under Governance
Snowflake and Databricks are where enterprise AI stops being a pilot. Orcaworks brings hands-on warehouse and lakehouse engineers — architects who build, not advisors who observe — together with the control plane that keeps every agent, model and query inside the policies your data already wrote.
Warehouse & lakehouse engineering
The unglamorous work that decides whether the AI layer holds up.
- Automated data discovery, ingestion and correlation across source systems
- Modelling on Iceberg and Delta, with Dynamic Tables and Lakeflow pipelines
- Warehouse sizing, workload isolation and consumption tuning
- Migration and consolidation across Snowflake, Databricks and legacy stacks
AI and ML built in the platform
Models and agents that live next to the data, not in a side system.
- Predictive models — churn, routing, demand, SLA and incident risk — trained and served in-platform
- Retrieval over governed enterprise content with Cortex Search and AI Search
- Natural-language analytics via Cortex Analyst and API/BI-class interfaces
- Agent build and evaluation on Cortex Agents, Agent Bricks and Mosaic AI
Governance that travels with the agent
The part most AI programs bolt on late, and pay for twice.
- Horizon Catalog and Unity Catalog policy enforced at execution time
- Authorized versus unauthorized action monitoring, with a hard stop
- Real-time cost and token attribution by team, agent and use case
- Column-level lineage from an AI answer back to its source
Scale without re-staffing
Built to move a proven use case across regions and business units.
- Blended onshore/offshore pods of Snowflake and Databricks practitioners
- Reusable patterns, runbooks and templates so region two costs less than region one
- Knowledge transfer to your internal team as the default, not an upsell
- Run-and-support models for solutions already in production
Weeks 1–2
Landscape and readiness
Catalog, lineage and policy baseline across your Snowflake and Databricks estate. Inventory of every model, agent and integration already running.
Weeks 3–6
One use case to production
A single governed workflow deployed in-platform — masking, row-level policy, evaluation and full audit trail in place from day one.
Weeks 7–12
Scale and hand-off
Replicate across regions and business units, wire in cost controls, and hand over the operating model and runbooks to run it.