Questions, answered.

Agami is the trust layer for your data. It sits between your AI tools (Claude, ChatGPT, Gemini, Cursor, custom agents, and any tool that supports MCP) and your data warehouse, so every question your team asks gets the same governed answer. The semantic model is auto-generated from your schema, vetted by your data team, and gated by your auth and security rules before any query runs.

A trust layer is the system that decides what a business question means before it gets answered. It owns your definitions (what ARR means, what counts as an active account, how regions are grouped), your access rules (who can see what and who is asking), and the routing to the right source. Without a trust layer, an LLM invents answers from raw SQL. With one, every assistant returns the same governed answer because they all go through the same vetted definitions.

BI tools are for analysts building dashboards. Agami is for any team asking business questions from any tool. The semantic layer in a BI product is locked to that BI product’s UI; Agami’s trust layer serves the same definitions to Claude, ChatGPT, Gemini, Cursor, custom agents, and any tool that supports MCP. You can keep your BI tool; Agami replaces or extends the semantic layer underneath.

A dbt semantic layer requires your team to author every metric, dimension, and join by hand and ship it through dbt Cloud. Agami auto-generates the first pass from your schema, your data team refines what matters, and the model stays in sync as the database evolves. Agami also integrates auth, row/column security, and surface integrations (Claude, ChatGPT, Cursor, and any tool that supports MCP) that dbt’s semantic layer does not.

No. If your team has already invested in a semantic model in LookML, dbt MetricFlow, Cube, or another semantic-layer tool, Agami can read those definitions instead of generating new ones. You keep your existing source of truth; Agami’s trust layer adds auth, row/column security, and AI-client surface integration on top. You don’t end up maintaining two parallel definitions of ARR.

Agami queries in place against your existing warehouse or database. Supported sources include Snowflake, Databricks, Postgres, MySQL, Redshift, BigQuery, and Oracle. You can connect a new warehouse in hours using your existing read replicas, roles, and credentials.

Agami never copies or replicates your source tables. Queries run in place against your warehouse using your own credentials and read replicas, and results are returned to you, not stored by Agami. To run the service, Agami keeps your semantic definitions, audit logs, and a record of the questions asked and the SQL generated, governed by your retention policy. In an in-tenancy deployment, that operational data stays inside your own cloud. See our Privacy Policy for details.

Agami auto-generates the first pass of the semantic model from your existing schema, including initial metric and dimension definitions. Your data team then reviews and approves every definition before it ships. The semantic model, not the LLM, picks the formula, grain, and joins for each question. As your schema evolves (columns dropped, tables added, new sources connected), the model adapts in place, and changes are reviewed the same way.

Agami exposes an MCP server (Model Context Protocol). Any assistant that speaks MCP (Claude, Claude Code, Cursor, custom agents) plugs in directly. For assistants without MCP support, Agami integrates through their tool/plugin APIs. The same governed answer comes back regardless of which assistant the question was asked in.

Yes. Agents and agentic workflows you build in LangChain, CrewAI, n8n, OpenClaw, or your own stack reach your data through Agami the same way an assistant does, over MCP or its tool APIs. They can query and act on their own, with no one approving each step, but every query still passes through the same vetted definitions, auth, and row/column security. The trust layer governs an autonomous agent exactly as it governs a person asking a question in Claude.

agami-core is the source-available distribution of Agami, the trust layer between AI and your data. It is fair-code, licensed under the Agami Functional Use License: free to self-host for your own team, with a commercial license required only to expose data to people outside your org. Run it locally in Claude Code, Cursor, or VS Code, or self-host the server on your own cloud; credentials and data never leave your infrastructure, no telemetry. Write your definitions in YAML and ask questions in plain English from any MCP client. The source is at github.com/AgamiAI/agami-core.

agami-core (the source-available, self-hosted distribution) is free to self-host for your team under the Agami Functional Use License; a commercial license applies only when you expose data outside your org. The managed platform, Hosted (our cloud) or On-prem (your environment), is quoted based on your team size and deployment requirements. Book a 30-minute conversation with the founders to discuss pricing for your scope.

The managed platform is built for enterprise-level compliance: role-based access control, row and column level security, audit logs, and SSO. On-prem deploys it inside your own VPC or datacenter with dedicated single-tenant isolation, priority support, and a setup engagement. Specific compliance certifications and customer security review workflows are covered in our security review and contract discussion. Email contact@agami.ai or book a call to talk through what your security team needs.

agami-core runs locally in minutes once you clone the repo. Connecting a new warehouse on the Hosted tier takes hours, not months, because Agami uses your existing read replicas, roles, and credentials. A typical Hosted deployment is live in days; On-prem deployments include a setup engagement and run on your timeline.

Agami was founded by Sandeep Kachru (Co-founder and CEO) and Ashwin Ramachandran (Co-founder and CPTO). Sandeep built data warehouses, semantic-layer systems, and eval pipelines at scale. Ashwin led product and engineering, turning the hardest business questions into trusted answers. Both ran data systems internally at large tech companies; Agami packages that pattern for everyone else.

The fastest way to see it work is to try the demo, which runs on real public data inside Claude or ChatGPT. To run it on your own data, talk to us: book a 30-minute conversation with the founders and we will set you up with your own data.