Leadership
Steer with the full picture.
Ask the strategic question and see the numbers, context, and evidence together. Give recurring decisions an agent that brings changes to your attention.
Less waiting for a report. More time to decide.
Agnes · Enterprise AI
Give your company a brain
Bring business data, company knowledge, and personal work context together—so every answer can lead to useful work.
SemanticsDefinitionsMemoryDecisionsKnowledgeDocuments
What becomes possible
Ask the question. Get the
answer. Put it to work.
Finance data, CRM records, contracts, and operating notes each hold one piece of the explanation.
AgnesCompany workspace
Alex takes the revenue question further
Why did revenue miss plan—and what should we change before it happens again?
Alex · COO · August revenue review
3
months as
the first user
If it could not carry my daily decisions, it was not worth shipping to you.
He ran his own daily decisions through Agnes, including a major partner negotiation, before asking anyone else to install it.
The problem is not a lack of data
Your company knows more than your AI does.
One business question. Disconnected sources, different definitions, and no shared explanation.
Scattered evidence
The answer starts in six different places.
Finance data, CRM records, contracts, and operating notes each hold one piece of the explanation.
Slackleadership · Today
Six sources. No shared explanation.
Conflicting meaning
Even “revenue” can mean different things.
Bookings, recognized revenue, and cash receipts tell different stories. The definition has to travel with the number.
Slackleadership · Today
Three reports. Still no agreed comparison.
Unsupported confidence
A plausible answer is not yet a defensible one.
Without the right definitions and sources, AI can produce a fluent explanation that nobody can verify.
Claude
Copied messages onlyRevenue review · Slack thread pasted
Why did revenue miss plan? Summarize this thread and suggest next steps.
Likely explanation
- Timing: two renewals moved; some signed contracts start next month.
- Different measures: bookings and cash receipts don’t equal this period’s revenue.
Next step: reconcile contract dates with recognized revenue before revising the forecast.
Which renewals? Show the source rows and their share of the shortfall.
I only have the pasted messages—not the reports or contract records. I can’t identify the renewals or calculate their impact. This is a hypothesis—not a supported finding.
Which definition?Which period?Which source?A useful hypothesis. Still no evidence to verify it.
Introducing Agnes
Give Agnes a job.
Not just a question.
Agnes connects your data, shared knowledge, and business definitions—so your teams and AI tools can work from the same foundation.
Work in the tools your team already knows
Talk to Agnes
Agnes web chat
Slack
Microsoft Teams
Telegram
Agnes web chatSlackMicrosoft TeamsTelegramChat & channels
Agnes web chat
Ask questions, inspect supporting sources, and create reports or drafts in the Agnes web application.
Slack
Use the configured Slack bot to ask Agnes a question and continue the work from your team’s channel.
Microsoft TeamsPlanned
A planned way to ask Agnes from team conversations. Today, meeting transcripts can be added as documents through supported ingestion.
Telegram
Use the configured Telegram bot for questions and follow-ups, connected to your Agnes instance.
Use your AI tools
Claude
Cursor
GitHub CopilotYour own tools
ClaudeCursorGitHub CopilotYour own toolsWith company context
Claude
Connect Claude to Agnes through MCP—the connection that lets your AI tool retrieve company context and use permitted tools. Claude produces the response.
Cursor
Connect Cursor to Agnes through MCP to query permitted data and retrieve relevant company knowledge while you develop.
GitHub Copilot
Use GitHub Copilot Chat in VS Code with the Agnes MCP connection. This is the GitHub Copilot integration, not Microsoft 365 Copilot.
Your own tools
Connect a compatible MCP client, or integrate scoped agents through the API to bring company context and configured work into your own application.
Build & analyzeClaude CodeInternal applications
Claude CodeInternal applicationsFrom insight to action
Analyst workspace
Use Claude Code with the Agnes CLI, curated data, and reusable tools for queries, notebooks, and recurring deliverables.
Internal applications
Build a maintained application on business data. Review its output and audience before sharing it with the team.
Agnes
One connected foundation. Eight building blocks.
Meaning & evidence
Semantic layer
Make every number mean the same thing.
Business glossaryMetric formulasData relationships
Connect business language to the data behind it. Agnes reads agreed definitions, formulas, filters, and relationships before querying, and can show which definition shaped the answer.
Business owners define and approve the meaning; query checks are advisory.
Documents & knowledge
Find the passage that answers the question.
Indexed sourcesSearchCitations
Bring connected documents into searchable collections. Agnes retrieves relevant passages from converted text and attaches their sources to the answer. Notes and transcripts can become knowledge when they are brought into an authorized collection.
Retrieval depends on what has been indexed and the access configured for the collection.
Structured data
Work with the facts in your business systems.
Warehouse queriesData packagesSQL analysis
Query registered datasets in supported warehouses or synchronize them on a schedule. Data packages bring together tables and the context needed for a business purpose, with configured access and query guardrails.
Data preparation, refresh frequency, and source support depend on the setup.
Knowledge graph
Connect the people, accounts, and facts behind an answer.
Business entitiesEvidence-backed linksConflicts
Connect entities through relationships supported by source passages. Explore which account, person, document, or business event belongs to the question, while keeping conflicting or incomplete evidence visible.
The graph reflects available evidence; it is not automatically complete company knowledge.
Memory & execution
Corporate memory
Make reviewed knowledge useful beyond one conversation.
DecisionsReview & approvalShared knowledge
Preserve useful decisions, procedures, and working knowledge after review. Distribute approved memory to relevant groups, revisit it, and withdraw it when it is no longer valid.
A discussion is not automatically an approved company decision.
Skills & plugins
Give AI your team’s methods and tools.
Reusable skillsPlugin libraryTeam playbooks
Package your team’s instructions into reusable skills. Organize and distribute them through plugins, so people and agents can work with the same approved methods.
Skills and plugins are reviewed and granted to the groups that may use them.
Agents
Give recurring work a dedicated agent.
Named agentsScoped accessScheduled work
Give an agent a specific job, the skills and tools it needs, and a schedule. It can investigate, compare, draft, or monitor a business question within its assigned scope.
Each agent has configured permissions, budgets, and review requirements.
Applications & artifacts
Turn an answer into something your team can use.
Reports & filesInternal applicationsDrafts & previews
Produce reports and downloadable files, or create applications that work with business data. Previews, logs, and deployment history help teams inspect and maintain the result.
Sharing an application publishes outputs under its owner’s data authority.
Grounded in your company context
Business systems
CRM · ERP · finance · operations
Structured data
Connect your warehouse directly. Bring operational systems through Keboola’s ingestion and preparation.
CRM and ERP connections depend on the Keboola connector and source setup.
Company knowledge
SharePoint · contracts · policies
Unstructured content
Turn company documents into searchable knowledge, with passages that support each answer.
Legacy Office files are converted first. Other sources use configured ingestion or adapters.
Your work context
Email · calendar · meeting notes
Personal context
Bring the context around your work into the question—not just the numbers behind it.
Use supported connections or uploads. Live connections depend on the integrations configured for your instance.
Agents at work
Turn recurring questions
into standing assignments.
Set the trigger. Your agent runs automatically and brings the work back for review.
Example automation
Starts automatically
Every weekday
Pipeline upkeep agent
Powered by Agnes
The assignment
Find stale deals. Prepare the CRM updates.
Arrives for your review
CRM updates ready for confirmation
You approve before changes are applied or sent.
Have a recurring job in mind? Let’s work through it together.
Get a consultationPerformance on real business questions
Better business answers.
Beyond AI with files.
Agnes connects company knowledge, business definitions, and the relationships behind the facts—so AI can handle questions that take more than finding the right document.
| 27graded answers per setup9 questions × 3 runsPassPartialFailOne dot per run | Connected company data and knowledgeAgnes27 / 279 of 9 passed all three runs | Document connectorChatGPT21 / 276 of 9 passed all three runs | Document connectorClaude19 / 274 of 9 passed all three runs | Documents and curated definitionsClaude + context16 / 275 of 9 passed all three runs |
|---|---|---|---|---|
| Precedent recall: | Agnes: pass, pass, passAgnes: pass, pass, pass | ChatGPT + SharePoint: pass, partial, passChatGPT + SharePoint: pass, partial, pass | Claude + SharePoint: pass, partial, partialClaude + SharePoint: pass, partial, partial | Claude + SharePoint + context: partial, partial, partialClaude + SharePoint + context: partial, partial, partial |
| Closest match: | Agnes: pass, pass, passAgnes: pass, pass, pass | ChatGPT + SharePoint: pass, pass, passChatGPT + SharePoint: pass, pass, pass | Claude + SharePoint: pass, pass, partialClaude + SharePoint: pass, pass, partial | Claude + SharePoint + context: fail, fail, passClaude + SharePoint + context: fail, fail, pass |
| Ambiguity: | Agnes: pass, pass, passAgnes: pass, pass, pass | ChatGPT + SharePoint: partial, partial, partialChatGPT + SharePoint: partial, partial, partial | Claude + SharePoint: fail, pass, partialClaude + SharePoint: fail, pass, partial | Claude + SharePoint + context: partial, partial, partialClaude + SharePoint + context: partial, partial, partial |
| Counting: | Agnes: pass, pass, passAgnes: pass, pass, pass | ChatGPT + SharePoint: pass, pass, passChatGPT + SharePoint: pass, pass, pass | Claude + SharePoint: pass, fail, passClaude + SharePoint: pass, fail, pass | Claude + SharePoint + context: partial, partial, partialClaude + SharePoint + context: partial, partial, partial |
| Entity resolution: | Agnes: pass, pass, passAgnes: pass, pass, pass | ChatGPT + SharePoint: fail, fail, passChatGPT + SharePoint: fail, fail, pass | Claude + SharePoint: fail, fail, passClaude + SharePoint: fail, fail, pass | Claude + SharePoint + context: pass, pass, passClaude + SharePoint + context: pass, pass, pass |
| Every setup passed every run | ||||
| Synthesis: | Agnes: pass, pass, passAgnes: pass, pass, pass | ChatGPT + SharePoint: pass, pass, passChatGPT + SharePoint: pass, pass, pass | Claude + SharePoint: pass, pass, passClaude + SharePoint: pass, pass, pass | Claude + SharePoint + context: pass, pass, passClaude + SharePoint + context: pass, pass, pass |
| Specific facts: | Agnes: pass, pass, passAgnes: pass, pass, pass | ChatGPT + SharePoint: pass, pass, passChatGPT + SharePoint: pass, pass, pass | Claude + SharePoint: pass, pass, passClaude + SharePoint: pass, pass, pass | Claude + SharePoint + context: pass, pass, passClaude + SharePoint + context: pass, pass, pass |
| Provenance: | Agnes: pass, pass, passAgnes: pass, pass, pass | ChatGPT + SharePoint: pass, pass, passChatGPT + SharePoint: pass, pass, pass | Claude + SharePoint: pass, pass, passClaude + SharePoint: pass, pass, pass | Claude + SharePoint + context: pass, pass, passClaude + SharePoint + context: pass, pass, pass |
| Negative case: | Agnes: pass, pass, passAgnes: pass, pass, pass | ChatGPT + SharePoint: pass, pass, passChatGPT + SharePoint: pass, pass, pass | Claude + SharePoint: pass, pass, passClaude + SharePoint: pass, pass, pass | Claude + SharePoint + context: pass, pass, passClaude + SharePoint + context: pass, pass, pass |
In a customer-run comparison, Agnes earned 27 passing answers out of 27, compared with 21 for ChatGPT, 19 for Claude, and 16 for Claude with added context. All three alternatives had access to company documents. The difference showed up in ambiguous questions, counting engagements, and connecting company relationships.
Early feedback from one customer’s evaluation, September 2026; anonymized. Nine document-based questions, three attempts per setup. Results reflect these configurations and their different data access, not a general ranking of the models. Three Agnes passes were qualified: it surfaced a source conflict without resolving it.
Built for your priorities
Move the business forward.
Keep control of how.
Finance
Make AI earn its place.
Track model spend, set agent budgets, and measure results against the work they replace. Give the business finance-approved numbers it can trace back to the source.
Accountable AI spend. Numbers you stand behind.
IT
Enable AI. Keep control.
Control who can access company context and which skills and agents they can use. Manage scope, review activity, and bring approved capabilities into compatible tools.
Useful AI for teams. Governance you can administer.
Before the first call
How Agnes fits your company.
How Agnes fits your tools, where it runs, and what your team needs to get started.
Fit and alternatives
Can we use Agnes with the AI tools we already have?
Yes. Agnes supplies shared business definitions, connected data, and reviewed knowledge to compatible AI tools through MCP, a standard connection for AI tools. Supported clients include Claude, ChatGPT, Cursor, and GitHub Copilot Chat in VS Code. This is not the same as a Microsoft 365 Copilot integration. Each client and its access are configured for your instance.How is this different from Snowflake Cortex or Databricks?
Agnes adds a shared layer of business definitions, documents, reviewed knowledge, and reusable methods across supported sources. It can use registered Snowflake and Databricks tables alongside company documents. The right combination depends on the sources, workflows, and controls you already have.Do we need the Keboola platform underneath it?
No. Agnes can run independently with its supported warehouse and document connections. Keboola adds ingestion, preparation, and orchestration when operational data needs to be collected or made ready for analysis. We scope that separately around the sources your use case needs.Deployment and security
Can our data stay inside our own network?
Agnes can run in your infrastructure, with a dedicated instance for your organization. Depending on the connection, data may be queried in place or copied into that instance. Model calls can send context to the configured external provider; local-model options apply to supported pipeline stages. Fully air-gapped operation is not currently supported. Deployment and data flows are agreed during scoping.Who can see what, and how is that administered?
Agnes supports configured Google or Microsoft sign-in and group-based grants for data and capabilities. It checks access server-side. Connected tools and channels must be configured and verified; source permissions are not automatically mirrored into Agnes. Admins manage who can use each collection, data package, skill, or agent.What actually reaches an external AI service?
The model may receive the prompt, retrieved passages, query results, and tool output used for the task. The exact data flow depends on the provider, tools, and pipeline configuration. Sanitization can be configured for supported document workflows. When you use an external AI client, its own data-handling settings also apply.Running it
What happens when it gets something wrong?
Your team reviews the answer and its sources to find the gap: a definition, missing data, outdated knowledge, or the instructions used. Assign an owner, correct the shared source or method, then retest the question. Agnes supports that improvement process; it does not remove the need for review.Is this employee monitoring?
Agnes is designed for business questions and workflows. It records usage and activity for cost control, troubleshooting, and oversight, and admins can review that activity. Define the scope and access for personal work context before connecting it; do not treat it as unrestricted access to employee communications.How long before something actually works?
The first engagement runs four to eight weeks, scoped to one executive and the data their real questions need. It ends with their decisions running through Agnes, the data packages those questions turned out to require, and an honest list of what it could not yet answer. The coverage map is yours whether or not you continue.
40 seats
By invitation onlyKeboola customers
Meet Agnes in person.
One morning in Prague with a small room of Keboola’s closest customers, and the first live look at the company brain.
- 15 October 2026 · 9:00 AM – 12:30 PM
- Keboola office, Prague