Compare
Agnes vs Snowflake, side by side
What each product does today, taken from its own documentation. Previews are marked as previews, and where the answer depends on your setup, we say so.
SnowflakeScorecard
14 questions, one verdict each
Every Agnes answer links to the changelog entry where it shipped.
- Warehouse sources: Agnes
- Data foundation: Agnes
- Access per person: Agnes
- Semantic layer: Agnes
- Owning definitions: Agnes
- Learning from usage: Even
- Where your definitions live: Agnes
- Document indexing: Agnes
- AI tools people already use: Even
- Model choice and your own key: Depends
- Agents that do the work: Agnes
- Sharing results: Agnes
- Where it runs: Agnes
- Definitions in outside tools: Depends
- Available
- Available, admin-controlled
- Preview or limited
- Not documented or not built
Showing 14 of 14 questions.
Data and sources
Warehouse sources
Can it use data that is not in Snowflake?
Live, zero-copy queries across every warehouse you have, not only Snowflake.
SnowflakeAvailable: Works on data in Snowflake. Other data is loaded into Snowflake first.
AgnesAgnes aheadAvailable: Live queries, zero copy: each question runs inside Snowflake, BigQuery or Databricks and only the answer comes back, so numbers are always current and nothing is duplicated. Keboola tables and scheduled copies when you want them.
Data foundation
What if half our data still sits in an ERP like SAP?
Your data does not have to move into one warehouse first.
SnowflakeAvailable: Openflow connectors and Snowflake's loading tools bring source systems into Snowflake.
AgnesAgnes aheadAvailable: Built on Keboola's data foundation: 1,500+ connectors, ELT and ETL, transformations, scheduled automation, and Kai, the AI agent that builds the pipelines with your team.
Access per person
Does each person see only what they are allowed to?
One permission model across every source and tool. Snowflake's own row policies are not mirrored; Agnes applies its own.
SnowflakeAvailable: Each person's Snowflake role and policies apply to the data inside Snowflake.
AgnesAgnes aheadAvailable: Group grants on every data package, table, plugin and agent, plus sharing for reports and apps. Row filtering and column masking, set per group by your admins, on live tables and on copies held on the server, Snowflake included. SharePoint document permissions mirrored, and BigQuery under each person's own account.
Definitions and semantics
Semantic layer
Do our agreed definitions reach every tool?
Snowflake stays the master for what it defines; Agnes reads it and adds the rest.
SnowflakeAvailable: Semantic views run natively inside Snowflake.
AgnesAgnes aheadAvailable: Several semantic models in one layer, with a business glossary, metric definitions and a Knowledge Graph. Reads Snowflake semantic views every 6 hours by default, alongside Databricks metric views, Keboola, git and uploads.
Owning definitions
Who approves what a business term means?
Governed definitions in production today; Snowflake's steward model is still in private preview.
SnowflakePreview or limited: Business Ontology: a steward per domain, draft to approved. Private preview.
AgnesAgnes aheadAvailable: Each definition has one source that owns it: git, Keboola or your warehouse. Imported models are read-only in Agnes, every correction waits for an admin's approval, and metrics can carry approvals and owners.
Learning from usage
Does it get better from how people ask?
EvenDifferent scope: Snowflake learns definitions, Agnes learns company knowledge.
SnowflakePreview or limited: Cortex Sense drafts context from query history, dbt and dashboards. Private preview.
AgnesAvailable, admin-controlled: Shared memory from conversations with a review queue, plus a record of what people correct. Does not read query history.
Where your definitions live
What can we take with us later?
SnowflakePreview or limited: Context features run in Snowflake. Ossie and RDF import and export is stated in a product blog, not yet in the documentation.
AgnesAgnes aheadAvailable: Snowflake stays the master, nothing has to move. Models in Agnes export as open Ossie YAML.
Documents
Document indexing
Can answers use our contracts, decks and notes?
Documents do not have to move into a warehouse first.
SnowflakeAvailable: Cortex Search over documents loaded into Snowflake. The SharePoint connector can carry file permissions.
AgnesAgnes aheadAvailable, admin-controlled: SharePoint and uploaded files, indexed where they live, with names pseudonymized before storage in the folders you choose and each document's permissions mirrored. A Knowledge Graph keeps each fact's source quote.
AI access and models
AI tools people already use
Can people ask from Claude, ChatGPT or their editor?
EvenAgnes also carries documents and knowledge kept outside Snowflake.
SnowflakeAvailable: A managed MCP server serves Cortex Analyst, Cortex Search and Cortex Agents to outside agents.
AgnesAvailable: Claude, ChatGPT, Cursor and VS Code over MCP, with the same data, definitions and skills, under each person's permissions.
Model choice and your own key
Whose AI model runs, and on whose contract?
DependsSnowflake for one bill. Agnes for your own key and model contract.
SnowflakeAvailable: Models from OpenAI, Anthropic, Meta, Mistral, DeepSeek and xAI that Snowflake runs, billed in Snowflake credits.
AgnesAvailable: Bring your own key: Claude on your Anthropic key or in your own Google Cloud through Vertex AI, or chat through your own AI gateway. Other AI work runs on OpenAI, Gemini or any OpenAI-compatible provider, routed task by task.
In the changelog
Agents and shared results
Agents that do the work
Can it run recurring jobs, not only answer?
SnowflakeAvailable: Cortex Agents combine SQL over semantic views with Cortex Search, set up by your data team.
AgnesAgnes aheadAvailable: Describe the job and the builder sets up the agent. Agents run on a schedule or when another system starts them, and colleagues can share them.
Sharing results
Can people share what they built from an answer?
SnowflakeAvailable: Streamlit apps in Snowflake, built by your data team.
AgnesAgnes aheadAvailable: Reports at a stable link for the right people, pages that keep their numbers current, and data apps from a chat request.
Running it
Where it runs
Who operates it, and where?
Your choice of who runs it and where, with no index shared with other companies.
SnowflakeAvailable: Inside your Snowflake account.
AgnesAgnes aheadAvailable: A single-tenant instance for your company only: managed for you on Keboola's Google Cloud infrastructure, or on your own servers in your cloud, AWS and Azure included.
Definitions in outside tools
Can an outside AI tool read the approved definitions?
DependsNeither reaches Business Ontology from outside today.
SnowflakeNot documented or not built: Not documented for Business Ontology. The MCP server lists no ontology tool.
AgnesAvailable: The semantic views Agnes reads reach every connected tool. Agnes does not read Business Ontology.
Where Agnes adds value
Where Agnes adds value for a Snowflake company
What Agnes brings to a company that already runs on Snowflake. Every card is in the released product; some are admin-controlled, so you decide who gets them and when.
- Available
Live queries, zero copy
Agnes asks your warehouse directly: each question runs inside Snowflake, and in BigQuery or Databricks if you have them, and only the answer comes back. Numbers are always current, there is no pipeline to keep in sync, and your warehouse's governance stays in force.
- Available, an admin connects it
Agnes reads your semantic views
Each view becomes a read-only Semantic Model in Agnes, and Snowflake stays the master. Comments, custom instructions and verified queries come across. It sees only what the connection's role can see.
- Available
The same context in the tools people already use
Claude, ChatGPT, Cursor and VS Code connect over MCP and get the same data, definitions and skills, under each person's Agnes permissions. Microsoft 365 Copilot is not on this list.
- Available
One semantic layer beyond Snowflake
Databricks metric views, Keboola, git or an upload feed the same layer, and models export as Ossie YAML. This matters only if you have more than Snowflake.
- Available
Runs on infrastructure your IT operates
Snowflake's context features run inside Snowflake. A Snowflake-native company's data is in Snowflake anyway, so this is rarely decisive.
- Available, admin-controlled
Semantic annotations
Reviewed caveats and corrections on a model Snowflake owns reach every reader in Agnes, and Snowflake is not changed. This is the closest fit to Snowflake staying the master.
- Available, admin-controlled
Metric governance
Approvals, owners and verified numbers for metrics. It covers metrics, not glossary terms.
- Available, admin-controlled
Shared memory and session analysis
Organisation knowledge learned from conversations, with a review queue, plus a record of what people correct or find missing. Snowflake's announced memory is per person. Both add AI model cost, and Cortex Sense's corrections cover part of this.
- Available, admin-controlled
Documents kept outside Snowflake, with a Knowledge Graph
SharePoint collections, and Claims that carry their source quote. Extraction adds AI model cost, so we scope it with you first, and we have not measured yet how much it improves answers.
- Available, admin-controlled
Row rules on live Snowflake tables
An admin writes these rules in Agnes, using Agnes identities and groups. Snowflake's own row policies are not mirrored.
Every item below is in the released product. Some are admin-controlled: your admins decide who gets them and when, which is part of governing AI across a company. Nothing here depends on moving your definitions out of Snowflake. If you do not need reach beyond Snowflake, you do not need this list.
Which should be the master
Which should be the master for your definitions
Each definition should have one master. For a Snowflake-native company that master can be Snowflake, and Agnes reads it instead of competing with it.
Snowflake is the master, Agnes reads, when
your metrics live in semantic views, or will, and your data team works in Snowflake. You have named owners for your terms. Agnes is there for reach beyond Snowflake and does not need to own definitions. Today the master Agnes can read is the semantic view, so keep what agents need in its descriptions and custom instructions.
Do not make the ontology the master yet, when
you need approved definitions in production now, because Snowflake says preview support is not production support. Or your definitions span warehouses or BI tools outside Snowflake, in which case keep them in git or in Agnes. Or nobody owns the terms yet: name owners first, because every tool depends on them.
Snowflake alone may be the better answer when
your questions stay inside Snowflake data, your people work in Snowflake's tools, and your documents can move into Snowflake.
The practical next step is to take a set of your own questions and run them both ways, then compare the answers.
Bring your questionsSnowflake today
What Snowflake gives you today
Weigh Agnes against the whole Snowflake stack: Cortex Agents, Cortex Search, Cortex Sense and Business Ontology together. Some of it is shipped and some is in preview.
- Shipped
A managed MCP server
Outside AI agents can use Cortex Analyst over semantic views, Cortex Search and Cortex Agents through Snowflake's own MCP server. Snowflake's release notes date it from November 2025.
- Shipped
Numbers and documents in one agent
Cortex Agents combine SQL over semantic views with Cortex Search over documents. The Openflow connector for SharePoint is generally available, and it can carry file permissions into Cortex Search.
- Private preview
Business Ontology
Business concepts with a steward for each domain, moving from draft to approved. Agents keep using the approved version until a steward accepts a change. It is open to selected accounts only, and Snowflake says preview support is not production support. Its documentation lists limits, for example that drift detection is not available.
- Private preview
Cortex Sense
It drafts context from query history, dbt models, dashboards and semantic views, and takes corrections through an inbox and evaluations. Documents in a stage are text and Markdown only for now, with PDF and Office listed as coming soon.
- Announced, no dates
Public preview and User Memory
Snowflake's product post says public preview of Business Ontology and Cortex Sense is coming soon. CoWork User Memory covers one person's role and preferences, and Snowflake describes shared knowledge as Cortex Sense's job.
- Stated in a blog, not in the documentation we read
Ossie and RDF
Snowflake's product post says Business Ontology can import and export Apache Ossie and RDF. The ontology documentation we read describes importing a file through Snowflake's tools and does not mention Ossie, RDF or export. We have not seen it working.
Where your data goes
You choose the AI provider
Every model call goes to the provider your deployment is configured to use. Pick one below to see who reads a question and whose contract it runs on.
A person asks
In Agnes, or in Claude, ChatGPT, Cursor or VS Code, under that person's permissions.
Your Agnes instance
Single-tenant, yours alone. It queries your warehouse live, with zero copy: only the answer comes back. Indexed documents sit here.
The model you chose
The question, with the data and passages found to answer it, goes to the provider you configured.
Anthropic APIThe answer comes back
With its sources, so people can check where each number came from.
- Who reads the question
- Anthropic reads each question to answer it.
- Whose contract
- Chat runs under your own Anthropic key, so the model relationship is yours.
We don't claim your data never leaves your cloud: the AI provider has to read a question to answer it.
If people use Agnes from Claude Code, Claude Code talks to its own model provider as usual, and the tables a person is allowed to see can be copied to their laptop for fast local analysis.
10 of 10 questions shown.
No. Agnes does not read Business Ontology today. It reads Snowflake semantic views, and Snowflake stays the master for what they define. The word ontology also means two different things here. Snowflake's is a governed glossary with relationships that people approve. In Agnes, an ontology defines what may be pulled out of documents into the Knowledge Graph. They are not the same thing, so do not compare them as one.
Yes. An admin connects your Snowflake account, and each semantic view becomes a read-only Semantic Model in Agnes. Comments, custom instructions and verified queries come across. It refreshes every 6 hours by default and sees only what the connection's role can see. A metric runs in Agnes only if its expression can be converted and it points at a table Agnes has registered; otherwise Agnes marks it not runnable. We prove this on your own Snowflake account against your own questions before we promise anything.
It doesn't have to. Agnes queries Snowflake live, with zero copy: each question runs inside Snowflake and only the answer comes back, so numbers are always current. If you want one, a scheduled copy is kept on your single-tenant Agnes instance and refreshed on a schedule, which makes repeated questions faster and spares Snowflake credits. Row rules an admin writes in Agnes can run inside Snowflake on live tables.
No. Agnes reads Snowflake through one connection and applies its own rules. If you need row rules on live Snowflake tables, an admin writes them in Agnes, using Agnes identities and groups, and Snowflake's own row policies are not mirrored. Inside Snowflake, access follows the person, and that is a real advantage of using Snowflake's tools directly.
Not as a general product yet. Business Ontology is in private preview for selected accounts, with no public date or price that we could find, and Snowflake says preview support is not production support. Ask your Snowflake account team about access and cost.
Snowflake hasn't documented direct access. Its managed MCP server, generally available, serves Cortex Analyst, Cortex Search and Cortex Agents, and its documentation does not list an ontology tool. An indirect route through a Cortex Agent may exist, but Snowflake's documentation warns that ontology search can return nothing through the agent REST API, and we have not tested it.
No. Agnes reads semantic views and does not write them. Agnes can import models in the open Ossie format from git or an upload, but we have not tested that against anything Snowflake exports.
Yes, once your admin switches it on for documents kept outside Snowflake. Extraction adds AI model cost, so we scope it with you first. We have not measured yet how much it improves answers.
When your questions stay inside Snowflake data, your people already work in Snowflake's tools, and your documents can move into Snowflake. Cortex Agents combine SQL over semantic views with Cortex Search over documents loaded there, and for documents already in Snowflake they do what we would offer.
Agnes facts on this page come from the Agnes product documentation and code, checked on 5 November 2026. Facts about Snowflake come from its public pages linked below. Snowflake's documentation for Business Ontology and Cortex Sense is limited-access, so where we rely on it we say "according to Snowflake's documentation". Business Ontology is in preview and Snowflake is still changing it, so check the current state with Snowflake. We found no independent hands-on review of it yet. If something here is out of date, tell us on the demo.
- Snowflake product post on Business Ontology (opens in a new tab) Three layers of context, and the statement that public preview is coming soon.
- Snowflake engineering post (opens in a new tab) How Business Ontology was built for AI agents.
- Snowflake-managed MCP server (opens in a new tab) Which Snowflake tools an outside agent can use.
- MCP server release note (opens in a new tab) Dated from November 2025.
- Cortex Agents (opens in a new tab) Agents that combine SQL over semantic views with document search.
- Openflow connector for SharePoint (opens in a new tab) Generally available, with a variant that indexes documents and their permissions into Cortex Search.
- Snowflake Cortex AI functions (opens in a new tab) The model families Snowflake runs, from OpenAI, Anthropic, Meta, Mistral AI, DeepSeek and xAI.
- Cortex cross-region inference (opens in a new tab) Model use is charged in Snowflake credits.
- Snowflake Openflow (opens in a new tab) Connectors that bring source systems into Snowflake.
- Streamlit in Snowflake (opens in a new tab) Apps built and run inside Snowflake.
- Snowflake CoWork (opens in a new tab) Where User Memory is described.
For how Agnes runs and where model calls go, see the deployment page.
AI Benchmark
See how Agnes did on real business questions
Cuesta Partners asked Agnes, ChatGPT and Claude the same 10 questions, three times each, and graded every answer themselves. One firm, one set of questions: directional, not a general ranking.
Share of graded answers that passed- Agnes100%
ChatGPT + SharePoint78%
Claude + SharePoint70%
Same 10 questions, 3 runs each, graded by Cuesta.