SQLore
FeaturesPricingDownloadAboutContact
Start freeStart for free
SQLore
HomeFeaturesSolutionsPricingDownloadCompareFAQAboutBlogContact

SQLore © 2026

TermsPrivacyRefunds
Download for macOSDownload for WindowsSign in
  1. Home
  2. Solutions
  3. Local models
Local models

A Local AI Database Assistant, Model Included

Local retrieval over your schema and a local model to write the SQL. Nothing about your database reaches anyone else’s server.

Download SQLoreSee all features

What is a local AI database assistant?

A local AI database assistant runs both the schema retrieval and the language model on your own hardware, so no part of your database — schema, question or rows — is sent to a third-party service. SQLore runs this way by default for PostgreSQL and MySQL; hosted models are an opt-in choice you make per connection.

Two things have to be local, not one

Tools describe themselves as local when the retrieval index sits on your machine while the model call still goes to a hosted API — which means your schema and your question leave your network on every request. Both halves matter. SQLore keeps the index local always, and with a local model configured, the generation step is local too, which is the point at which "private" stops being a policy and becomes a network fact you can verify with a packet capture.

What you give up, honestly

A local model is smaller than a frontier hosted one, and on the hardest questions — a five-table join with a window function over an inconsistently named schema — that shows. The trade is real and worth stating plainly rather than talking around.

  • Most day-to-day questions are handled locally without a noticeable difference.
  • Very large schemas and unusually complex analytical questions are where hosted models still lead.
  • Local inference speed is a property of your machine — good hardware is genuinely fast, modest hardware is not.
  • Retrieval quality is unaffected: the schema index is local either way, and it is what drives correctness.

Mixing the two is the practical setup

Stay local by default and reach for a hosted model on the questions that need one. Because it is a per-connection choice, the database with customer data in it can be local-only while the staging one is not — which is a policy you can actually enforce, rather than a habit you have to maintain.

Local does not mean unverified

The same guarantees apply whichever model wrote the query: parsed to an AST, validated to a single read-only SELECT, executed on a read-only connection under a statement timeout, and shown to you in full with the answer. A local model is not a more trusted model — it is a more private one.

Frequently asked questions

What hardware do I need to run a local model?

A current Mac with Apple silicon or a PC with a reasonably modern GPU runs a useful local model comfortably. Inference speed scales with the machine, so a slower one still works but answers take longer. The schema retrieval side is cheap on any hardware.

Is a local model as accurate as a hosted one?

Not on the hardest questions. Correctness in this setting is driven mostly by schema grounding rather than raw model size, and the retrieval index is local either way — so the gap is much smaller than the parameter counts suggest, and it shows up on complex multi-table analytics rather than on everyday questions.

Can I switch between local and hosted models?

Yes, per connection. That lets you keep sensitive databases local-only while allowing hosted models elsewhere, and it makes the choice an enforceable setting rather than something a user has to remember.

More answers on the main FAQ, or ask us directly.

Related

Air-gapped & regulated

An Offline AI Database Tool For Networks That Do Not Leave

Read more

The category, explained

The AI Database Assistant That Runs On Your Machine

Read more

PostgreSQL

Chat With Your PostgreSQL Database

Read more

Every page in this section covers a different question about the same product. See all solutions.

SQLore platform

Stop Writing The
Same Three Joins

Start free, upgrade when you are asking enough questions for it to matter.

Start for freeContact sales
sqlore.com
Connectionretail_db

staging

analytics

Add connection

MENU

Chat

Reports

Schema

History

Settings

Rows scanned

15,600+18%

Questions

25,800+14%

Avg latency

41 ms−16%

By product

By month

By connection