Step 01
Connect & discover
Point Agents at your SQL database. It maps tables, keys and relationships remotely — your rows never leave your infrastructure.
A secure MCP layer between AI agents and your database. Every agent gets access to exactly the rows and columns it’s authorized for — and every query it sends is on the record. Live data. Zero leakage.
Agents sits between AI and your data. You bring the agents — assistants your teams already use, or ones you build for your customers. We give each of them a scoped MCP endpoint to your database, and keep a record of every query they send.
Every agent connects to your database through the same MCP gateway — and each one can only reach the rows and columns it has been authorized for.
Off-the-shelf assistants or agents you build yourself — anything that speaks MCP plugs into one endpoint, with its own credentials.
Each agent’s row- and column-level rules are compiled into the SQL before anything runs. A prompt can’t widen them.
Which agent asked, the exact SQL that ran and what came back — from live data in place, with nothing copied.
Every team wants its AI agents connected to company data. Today that usually means handing an agent a database password and hoping for the best.
Agents connect with a broad service account. What each one may see is left to a prompt — and prompts can be talked around.
Which agent queried what, and when? Without a record, security and compliance teams have no choice but to say no.
Exports, vector stores, giant context windows. Every copy is another place to leak — and every row stuffed into a prompt is tokens you pay for.
No migrations, no data pipelines, no new chat interface to adopt. Agents sits on top of the database you already run and works with the agents you already use.
Point Agents at your SQL database. It maps tables, keys and relationships remotely — your rows never leave your infrastructure.
Decide which rows and columns every agent may reach. Agents compiles those rules into every query — not as a suggestion to the model, but as SQL.
Add the Agents MCP endpoint to any agent. It gets tools to explore the schema and run scoped queries — never the database credentials themselves.
Agents reads your database’s structure, never its contents. Tables, keys and relationships become the map every connected agent queries through.
The schema map is also where you decide who owns which rows and which columns are sensitive. Every query an agent sends is bound by it.
Three agents send the exact same query through the gateway. Each gets a correct answer — scoped to exactly what that agent is authorized to see. Pick an agent.
| ACCOUNT | REGION | REVENUE | MARGIN | CONTACT |
|---|
One place to follow everything your agents ask: which agent sent it, the exact SQL that ran after policies were applied, and what came back.
| TIME | AGENT | QUERY (AS EXECUTED) | SCOPE APPLIED | ROWS | STATUS |
|---|---|---|---|---|---|
| 09:42:18 | Sales assistantteam: sales-north | SELECT name, stage, amount FROM deals WHERE close_date < '2026-11-01' AND region = 'North' | region = North | 12 | ● Scoped |
| 09:41:03 | Finance agentteam: finance | SELECT product_line, SUM(margin) FROM orders GROUP BY product_line | contact masked | 8 | ● Masked |
| 09:39:47 | Customer portaltenant: acme-retail | SELECT COUNT(*) FROM orders WHERE tenant_id = 'acme-retail' | tenant = acme-retail | 1 | ● Scoped |
| 09:37:12 | Sales assistantteam: sales-north | SELECT * FROM salaries | table not in scope | 0 | ● Denied |
| 09:35:55 | Customer portaltenant: northwind | SELECT sku, SUM(qty) FROM order_items WHERE tenant_id = 'northwind' GROUP BY sku | tenant = northwind | 24 | ● Scoped |
Every query is tied to the agent that sent it, so you always know where a request came from.
See the final SQL after row and column policies were compiled in — not just what the model intended.
Rows returned, columns masked, requests denied. Evidence your security and compliance teams can work with.
Your agent decides what to ask. The Agents gateway decides what can actually run — deterministically, per agent. Your database stays where it is.
Schema discovery is remote. Nothing is exported, indexed or copied to answer a question.
Access rules are compiled into the SQL itself, so a clever prompt can't widen what a query returns.
No new interface to adopt. Agents works with the AI tools your teams and customers already use, over MCP.
Which agent asked, what ran, what came back — a full log for security and compliance teams.
The model sees a scoped schema, not your data. Aggregation happens in the database, so answers stay accurate and inference stays cheap.
Agents never receive database passwords. Each one gets its own MCP access that you can scope or revoke at any time.
Anywhere an agent needs data that not everyone should see, Agents fits.
A sales assistant answers questions about a team’s own accounts and deals — and never reaches another territory’s book.
A finance agent runs variance checks and cohort cuts straight from the ledger, with customer PII masked.
An operations agent checks inventory and SLAs for a single site, scoped to that site’s rows.
Give each customer’s agent MCP access to your product data — scoped to that tenant’s rows in a shared database.
Agents are everywhere, and MCP gave them a standard way to reach tools and data. What's missing is the layer that makes it safe to point them at a real database.
Teams already work with AI assistants every day, and MCP has made connecting them to data a configuration step. Writing the query is no longer the hard part.
Security and compliance teams block AI data access because they can't see or prove what agents queried. Agents logs every query, by agent.
Regulation and customer contracts increasingly forbid copying data into third-party AI stacks. Remote discovery means nothing has to move.
When access is safe by construction, every agent can get data — internal assistants for every team, and customer-facing agents for every tenant.
| Direct DB credentials | Generic text-to-SQL tools | AI Peri Agents | |
|---|---|---|---|
| Works with any MCP agent | Varies | — Own chat interface | ✓ One MCP endpoint |
| Row & column scope per agent | — All or nothing | — One service account | ✓ Compiled into every query |
| Log of every query, by agent | — | Limited | ✓ Agent · SQL · rows |
| Data stays in your database | ✓ | Varies | ✓ Remote schema discovery |
| Token cost per answer | High · rows into context | High · rows into context | Low · scoped schema, in-database compute |
| Safe to give customers’ agents | — | — | ✓ Tenant isolation |
We're working closely with a small group of early customers to shape Agents. Connect a database, plug in your agents, and see every query they send.
Hands-on onboarding, policy setup with our engineers, and direct input into the roadmap.
We'd love to walk you through the product, the architecture and where it's going. Write to us →