Insight O' Mate
Local-first NLP for NoSQL databases

Ask your database. Privately.

Insight O' Mate turns plain English into MongoDB queries — on your machine. The prompt travels. The data never does.

macOS · Windows · Linux

A different contract with your data.

Most NL-to-SQL tools want a copy of your database. Insight O' Mate is built so that was never an option in the first place.

  1. 01

    The cloud sees your question. Never your data.

    Only the prompt and your collection / field names travel to our stateless NLP service. Documents stay on the hardware you control.

  2. 02

    Bring your own database. We bring the language model.

    Point Insight O' Mate at any MongoDB you can connect to — local, self-hosted, or Atlas. The query runs on your side of the wire.

  3. 03

    No prompts logged. No training. No fine print.

    The analysis endpoint is stateless by design. Nothing is retained, nothing is replayed, nothing becomes tomorrow's model weights.

Live preview

One prompt in. A real query out.

Pick a question. Watch it compile into a concrete MongoDB aggregation pipeline and a result shape — the same payload the desktop app runs locally.

>

top 5 customers by revenue this month

POST /api/analyze
Analysisstateless
intent
aggregate.rank
collection
orders
fields
customerIdtotalcreatedAt
[
  { $match: { createdAt: { $gte: ISODate("2026-05-01") } } },
  { $group: { _id: "$customerId", revenue: { $sum: "$total" } } },
  { $sort:  { revenue: -1 } },
  { $limit: 5 }
]
Result · rendered locallyyour machine
customerrevenue
Aster Labs48210
Beacon Mfg.39804
Clarion Health31500
Dune Supply Co.28770
Everleigh Foods24995

Sample data. In the real app this executes against your MongoDB — nothing is uploaded.

Architecture

The shape of the wire is the promise.

Other NL-to-DB tools wave at “enterprise-grade security.” We show you the structure. Here is exactly what travels, and exactly what doesn't.

Your machine
local-first

Desktop app · sidecar · MongoDB

  • React frontend
  • Local Node sidecar
  • queryRunner.js
  • Your MongoDB

Raw documents are loaded, filtered, and aggregated here. The sidecar owns the DB connection string. Nothing in this box is ever transmitted during normal operation.

↗ prompt + fields
↙ analysis
Our cloud
stateless

nlp-engine · Cloud Run

  • preprocessing.js
  • nlpAnalysis.js

No database connection. No prompt log. No retention. The endpoint parses the prompt into an analysis object and returns it.

Sent upstream
prompt + metadata only
  • Prompt text
  • Collection names
  • Field names
Never sent
stays on your hardware
  • Document values
  • Database credentials
  • Connection strings
  • Query results
Received back
structured analysis
  • Intent classification
  • Target fields
  • Aggregation plan

Capabilities

Think local. Move fast.

Built for founder-operators who want an answer in seconds and the option to keep working on a plane, in a cafe, or behind a strict VPN.

offline-first

Analyses run on a local query engine. Your DB never leaves your LAN.

The desktop app talks to your database directly. Cloud Run only sees the prompt and your schema names — everything else, including all document values, results, and credentials, stays inside your network.

// packages/desktop-app/sidecar
→ POST /analyze            (prompt, fields)
← analysis                 (intent, pipeline)
→ collection.aggregate()   (on localhost)
← results                  (never uploaded)
auth

Keys live in the OS keychain.

The API key is stored in Keychain (macOS), Credential Manager (Windows), or libsecret (Linux). Never on disk in plaintext.

installers

Native installers for all three.

macOS (.dmg), Windows (.exe), and Linux (.AppImage + .deb). One Electron shell, same React frontend as the web.

< 0ms
typical analysis latency
0
bytes of your data uploaded
Free
for 10 questions / day
MIT
desktop app core

Get started

Point it at your database. Ask a question.

Free forever for one founder, one database, 10 questions a day.No credit card required.

no data uploaded·macOS · Windows · Linux·cancel anytime