A real database your AI agent can run.
Give the agent you already use — Claude Desktop, Hermes Agent, or OpenClaw — a real, private database it runs over MCP: flexible, encrypted, searchable, and yours. It files your receipts, totals your spending, hands you a report — things a chat can't. Underneath it's a serious database: full SQL, ACID, on your own encrypted volume, scaling to 250M records for the software you build. No AI in the box — the intelligence is your agent's.
Bring your AI. Write SQL. Administer the rest.
Every InventDB Serverless instance gives you four ways to work the same private database — the one you bring (your AI agent, over MCP), plus three built in: a SQL Studio to write and run SQL by hand, a Files browser to search every attachment by name, by the text inside it, or by meaning, and an Engine to administer it. The three built-in surfaces run no AI — the intelligence is whatever agent you point at it.
Bring your own AI, over MCP
The surface you bring. Point Claude Desktop, Hermes Agent, or OpenClaw at your database over MCP, and it reads, writes, searches and reports through 51 safe, typed tools — bounded by your role and row rules, every call logged. Below: Claude Desktop rendering a real owner statement straight from the ledger.
Bring your own:the intelligence is your agent's — InventDB ships no AI in the box; it's the private database the agent runs.
Write and run SQL by hand
A workbench, not a chatbot. Browse the schema of your namespaces, types and columns, write SQL in the editor and run it — a results grid, query history, and Download CSV / JSON.
AI-free:no AI assistant in this build; the /sql endpoint is open for your own tools and for external AI agents over MCP.
Find any file, one place
Every attachment across all your namespaces and types in one drive. Walk the folder tree, or search by filename, by the text inside a file, or by meaning — full-text and semantic search over PDFs, images and documents, with OCR built in.
AI-free:text and semantic search run on the OCR and embedding models built into the database — no external AI needed.
Administer the database
The admin surface: Relationships, Data Quality, Search Freshness, Attachment Processing, Team & Access, For Developers, Safekeeping & Health and Settings — everything to run the database, nothing it reasons over for you.
AI-free:no AI Budget and no Connections in this build — the data plane, with a clean way to run it.
For your AI agent — and the builders behind it. Not a business app.
Point the agent you already run — Claude Desktop, Hermes Agent, or OpenClaw — at a database it drives over MCP; or build straight against SQL and REST. A self-running database you own, not an AI that runs your business.
The DB behind AI-built apps
Spin up a database per app or per agent in seconds, then grow to production scale without a migration.
A serious backend
A real SQL database with ACID guarantees, full consistency, search and vectors built in, behind a stable URL with a REST API.
Your own encrypted volume
Single-tenant isolation enforced at the storage layer — your key, your volume, your process. Not a row in someone else's big table.
A few thousand rows today. 250 million later.
Same database, same SQL, same API — from your first prototype to a quarter-billion records. No painful migration from a prototype database to a production one.
Flat write throughput
Data is written in sealed, immutable segments with per-segment indexes — so writes stay fast as the database grows.
Skips what it can
Bloom filters and zone-maps prune whole segments before reading, so a query over 250M rows touches a tiny fraction.
Fast analytics at scale
Aggregations and GROUP BY run over compact columnar files, not row scans — dashboards stay fast at scale.
Everything you expect from a database. And more, built in.
InventDB Serverless prices, but a genuinely serious database underneath — with search, vectors and files you'd normally bolt on as three more services.
Full SQL
Real SELECTs with joins, filters, GROUP BY and aggregates — the query language you already know.
Crash-safe writes
Multi-record writes are all-or-nothing. WAL-backed and crash-safe, with self-healing corruption recovery.
Read your own writes
Fully consistent — no eventual-consistency window to design around.
Four ways to search
Exact, fuzzy (typos), full-text and semantic-by-meaning — all from one query layer.
Vectors built in
On-device embeddings and per-segment vector indexes. Build similarity search with no extra service.
Files with OCR
Store PDFs and images alongside records; the text inside is extracted and indexed for search.
Your own key
AES-256-GCM at rest with a per-database key you rotate or revoke. Every key access audited.
Schema without migrations
Add fields without downtime. The schema evolves transactionally as your app does.
MCP server built in
Every database ships a 51-tool MCP server — so any AI client can read and write it. No AI to host yourself.
One line to connect. Bounded by design.
Adding your database to Claude Desktop or any MCP client is a single line (right). From there, the 51 tools run bounded by the connecting credential's role and row rules — searching, safely writing (with review and audit), reading documents, running reports — with every call logged. The same guardrails apply to every agent, automatically, because row-level security and audit live in the database, not the agent. There's no chatbot in the box; that's InventDB SOAR's job.
{
"mcpServers": {
"mydb": {
"url": "https://<your-db>.inventdb.cloud/mcp",
"auth": { "type": "oauth2" }
}
}
}
▸ connected · 51 tools · bounded by your role
Connect however you build.
Just HTTP and SQL — send a query with a bearer token, get JSON back. Every interface runs bounded by the caller's role and row rules.
customers ▸ [ you · acme · globex · … ]
one big table, row filters
InventDB Serverless:
you ▸ /vol/you · own encrypted volume
acme ▸ /vol/acme · own encrypted volume
globex▸ /vol/globex · own encrypted volume
one writer per volume — enforced by storage
Not a row in someone else's big table.
Most cheap serverless databases pack everyone into one shared cluster and keep tenants apart with row filters. One bad filter, one bug, and your data is in someone else's query. We give every customer their own encrypted volume and their own database process.
- Dedicated encrypted volume per customer — no shared data plane
- One writer, guaranteed by the storage layer, not a config flag
- No noisy neighbors — isolation you can explain to a security reviewer
A database. Nothing more, nothing less.
InventDB Serverless is the pure database — it never reasons over your data and never implies built-in AI. You pay for the database; the intelligence is yours to bring.
Spin up a database in minutes.
Flat pricing from $11/mo, grows to 250M records on the same database. Bring your own AI over MCP, or build straight against SQL and REST.