Skip to content

Works with Claude, ChatGPT, Microsoft 365 Copilot and any MCP client

Enterprise Context for every AI agent.

Nexatron sits between your AI agents and your business systems, so every agent works from the same connected data, approved definitions, business rules and permissions.

  • SOC 2 Type I
  • ISO 27001 Certified
  • GDPR Aligned
  • SSO and SCIM

Our partner ecosystem

  • Anthropic
  • Databricks
  • Snowflake
  • Google Cloud
  • Microsoft
  • Salesforce
  • Oracle
  • NetSuite
  • IBM
  • UiPath
  • Workato

Inside the layer

Four pillars behind every answer.

Connected data brings the figures from where they live. Semantics & ontology says what they mean. Business context applies your rules. Governance decides who sees what and keeps the record.

Illustrative example with sample figures.

Connected data · Queried in place, 1 of 4

Straight from your systems, with no warehouse to load.

Queries run in your systems with user-level access. There is no warehouse to load. Nexatron keeps what it needs to answer and show its work, inside your tenant: the catalog, value lists from tables that aren’t row-secured, and the answers and rows people save or receive.

Illustration: the $42.8 million answer traced through the approved definition to revenue lines in the NetSuite general ledger for July to September 2026. Nexatron connects to 279 systems: 200 systems of record such as Salesforce, SAP S/4HANA, Oracle Fusion, NetSuite and ServiceNow; 30 data platforms such as Snowflake, Databricks, BigQuery and Amazon Redshift; 11 BI and semantic model tools such as Power BI, Tableau, Looker and dbt; and 38 docs and knowledge systems such as SharePoint, Google Drive, Confluence and Notion.

Semantics & ontology · Approved definitions, 2 of 4

Your definitions, not a model’s guess.

Definitions say what a metric counts, which entities it covers and how they connect across your systems. Every answer shows its trust level: approved by an owner, or interpreted and labeled so.

Illustration: the approved definition of recognized revenue, version 14, owned by Finance: revenue earned in the period, net of credits and intercompany eliminations, from the NetSuite general ledger. Below it, how the business connects: a customer, held in Salesforce and NetSuite, signs a contract in Salesforce; the contract is billed as invoices in NetSuite, and invoices post to ledger lines in NetSuite. Bookings, owned by Revenue Operations, counts contracts. Recognized revenue, approved by Finance, counts ledger lines. Agents follow these links, so bookings and recognized revenue never get mixed up.

Business context · Rules and calendars, 3 of 4

Your calendar, your close, your eliminations.

Quarters follow your calendar, figures stay labeled preliminary until the books close, and intercompany revenue is eliminated the way Finance does it. Nexatron learns these rules from your docs, dashboards and queries, and owners approve them.

Illustration: Q3 2026 runs from July to September. The books closed on October 12, so Q3 figures are final; before the close they were labeled preliminary. Intercompany revenue is eliminated on consolidation, and the answer is compared with the Q3 plan of $44.2 million.

Governance · Access and audit, 4 of 4

Answers that respect who is asking.

Row and column rules follow the person, whether they ask directly or through an agent. Anything outside their access never appears in an answer, and every request is on record.

Illustration: an access table. The CFO sees revenue for all four regions and the cost fields. The regional VP for APAC sees APAC only. An account executive in Northern Europe sees their own accounts, and so does ChatGPT acting for them. When that executive asks ChatGPT what bookings were in Q3, the answer is $2.4 million, for their 14 accounts; accounts outside their access never reach the answer. Every request is on record: who asked, through which agent, and what they saw.

Every agent, one answer.

Definitions, rules and permissions live in Nexatron, so Claude, ChatGPT, Microsoft 365 Copilot, Slack, Teams, your own agents and your apps all answer from the same context.

The questions your leaders ask every week.

Pick a leader and a question. Every answer shows its trust level, the system it came from, the definition it used, the rule it applied and what the asker can see.

The same approved answer comes back in any agent.

Claude, for the CFOApproved

What was recognized revenue against plan in Q3?

$42.8M3.2% below the Q3 plan of $44.2M

How Nexatron answered

Connected data
NetSuite general ledger
Semantics & ontology
Recognized revenue, v14
Business context
Q3 closed Oct 12: final
Governance
CFO: all regions

Illustrative example with sample figures.

Ask the way you work.

Chat, dashboards, agents and your own product all read the same approved definitions, so a number never changes with the tool.

Conversational BI. Ask in your own words. Every answer comes with the query behind it and the rows it returned. Example conversation. Asked how revenue compared to plan in Q3, Nexatron answers $42.8 million in recognized revenue, 3.2% below the Q3 plan of $44.2 million, approved, with revenue by region against plan: North America $18.9 million against $19.3 million, Northern Europe $8.1 million against $8.9 million, Southern Europe $7.9 million against $8.0 million, APAC $7.9 million against $8.0 million. The answer cites the NetSuite general ledger and recognized revenue version 14, with the SQL and lineage. Asked which region drove the gap: Northern Europe, 9.0% below plan; the other three regions were within 2.1% of plan.

Illustrative example with sample figures.

Three steps, no data project.

No migration and no new warehouse. Nexatron works with the systems and permissions you already have.

  1. Connect what you already run

    Queries run where the data is, under the permissions it already has. Most of the work is connecting and approving definitions, not moving data.

    Salesforce, NetSuite, SAP S/4HANA, Oracle Fusion, Snowflake, Databricks, BigQuery, ServiceNow and 271 more.

  2. Nexatron works out what it means

    It reads your schemas, profiles and query history, imports definitions you already have, and proposes definitions, joins and rules. Owners approve them, and every version is kept.

  3. Ask in your own words

    In text, by voice or through your AI assistant. Every answer shows its work.

    Claude, ChatGPT, Microsoft 365 Copilot, Slack, Teams and voice.

See what your teams can ask of the systems you already run, in Claude, ChatGPT and Microsoft 365 Copilot. Every connector is available today.

All connectors

AI assistants follow the same rules as people.

Every question, from a person or an agent, is checked against the requesting user’s permissions when it runs.

  • SOC 2 Type I
  • ISO 27001 Certified
  • GDPR Aligned
Visit the Trust Center
The regional VP for APAC asks Microsoft 365 Copilot“What was recognized revenue in Q3?”
  1. Who is askingSigned in through SSO. Copilot acts for the regional VP and gets their access, never more.
    Passed
  2. What they can seeRow rules: APAC only. Column rules: no cost fields.
    Passed
  3. What the query may doThe generated query is checked against policy before it runs, and redaction applies to the result.
    Passed
  4. On recordWho asked, through which agent and what they saw, in a tamper-evident log.
    Passed
$7.9MAPAC only, approvedRecognized revenue, v14. Q3 closed October 12.

Illustrative example with sample figures.

Isolation and access

  • Tenant isolation, checked on every request
  • SSO with OIDC, and SCIM provisioning
  • MFA, with step-up for data erasure
  • User-level access that inherits each source system’s permissions

Governance

  • Row and column-level security
  • Policy-aware redaction
  • A tamper-evident, hash-chained record of questions from people and assistants, plus administrator actions
  • Signed exports

Data protection

  • Encrypted in transit and at rest; warehouse credentials with AES‑256‑GCM
  • Model output validated before any query runs
  • Never used to train foundation models

Deployment

  • SaaS on AWS, with single-tenant VPC and on‑premises deployments on request
  • In your own deployment, use your own Anthropic or Amazon Bedrock account
  • SOC 2 report and DPA available under NDA

Any MCP agent becomes a governed analyst.

Add Nexatron as a hosted MCP server. Agents get the same approved definitions, permissions and audit log as the people they act for.

  1. Add Nexatron

    From Claude’s connector directory, or as a custom connector in any client that supports remote MCP servers.

  2. Sign in

    With your Nexatron identity, over OAuth 2.1 with PKCE. Clients that support it register themselves.

  3. Ask

    The agent calls Nexatron’s analytics tools. A workspace admin decides whether action tools are on.

  • Streamable HTTP
  • JSON-RPC 2.0
  • OAuth 2.1 with PKCE
  • Dynamic client registration
Hosted MCP serverStreamable HTTP

https://api.nexatron.io/api/v1/mcp

Analytics tools

  • ask
  • list_sources
  • list_metrics
  • explain_metric
  • query_metric
  • compare_metric_across_sources
  • audit

Action tools an admin can turn these off

  • list_actions
  • dry_run
  • execute_action

Works with Claude, ChatGPT, Microsoft 365 Copilot and any MCP client that supports remote servers.

Agent session · connected to Nexatron

Which region missed plan by the most in Q3?

query_metricrecognized_revenue, by region, Q3 2026

Northern Europe: $8.1M against a plan of $8.9M.

  • Approved
  • Row-level security
  • Logged

Claude is a trademark of Anthropic, PBC.

Common questions

More in the documentation and the Trust Center.

What is Nexatron?

An enterprise context platform. It gives people and AI agents the same approved definitions, business rules and permissions, so the answers match wherever they are asked.

What is a semantic layer?

Shared definitions of your metrics, joins and rules. Nexatron proposes definitions from your systems, and your owners approve them. Every change is versioned.

Which assistants can connect over MCP?

Claude, ChatGPT, Microsoft 365 Copilot and any MCP client that supports remote servers, over Streamable HTTP and JSON-RPC 2.0, with OAuth 2.1, PKCE and Dynamic Client Registration.

Which data sources does it support?

Warehouses and databases such as Snowflake, BigQuery, Amazon Redshift, Databricks, PostgreSQL, MySQL, SQL Server and DuckDB, files in S3 and Parquet, and business apps such as Salesforce, NetSuite, QuickBooks, ServiceNow, HubSpot, Stripe, Microsoft Dynamics 365 and Atlassian. Users sign in with their own account in each business app (or your SSO), and Nexatron inherits their existing permissions.

How is our data kept separate?

Every request is limited to your tenant and checked against row-level security before it runs. Warehouse credentials are encrypted with AES-256-GCM, and your data is never used to train foundation models.

What comes with an answer?

The result rows, the query that produced them, the tables it read, its trust level, and who on your team reviewed the logic behind it and when, once someone has.

Do people and agents get the same answers?

Yes. Both use the same approved definitions, access rules and audit log.

Give every agent
the same context.

Tell us which systems you run, and we plan a pilot on them with you.