🇺🇸 Glancer — Ask your Rails database questions in plain language

A Rails engine that allows you to query the database using natural language, with RAG and LLMs. No need to write any code.

Every Rails project reaches a point where someone on the team (product, support, management) needs data, but no one has time to write a quick query for them. You know the drill: a Slack message, a GitHub issue titled "how many users signed up last week?", or a CSV request that ends up in someone's backlog.

So, I think about Glancer.

What it is

Glancer is a Ruby on Rails engine that mounts a full chat interface at /glancer inside your app. You type a question in plain language, it retrieves the relevant schema context, generates a SELECT query (or an ActiveRecord expression), validates and executes it safely, then returns the result with a human-readable explanation.

"How many orders were placed in the last 30 days, grouped by status?"
→ SELECT executed, results shown, answer written in plain language.

It is inspired by Blazer, but instead of writing SQL yourself, you just ask.

Getting started

Add it to your Gemfile:

gem "glancer"

Run the generator:

rails generate glancer:install
rails db:migrate
rails glancer:index:all

Then visit /glancer and start asking questions.

The generator creates an initializer where you configure your LLM provider. A minimal setup with Gemini looks like this:

Glancer.configure do |config|
  config.llm_provider   = :gemini
  config.llm_model      = "gemini-2.0-flash"
  config.gemini_api_key = ENV["GEMINI_API_KEY"]

  config.schema_permission = true
end

OpenAI and OpenRouter are also supported. It runs on top of RubyLLM, so any model it supports works here too. You can assign different models per role: a smarter model for query generation, a cheaper one for writing the response.

What it can do

A few things worth knowing:

Safety is not optional. Every query runs inside a transaction that always rolls back. The keyword blocklist rejects DELETE, UPDATE, INSERT, DROP, and friends before they even touch the executor. You can also point it at a read-only replica.

No extra infrastructure. Embeddings live in your existing database as a JSON column. No Pinecone, no Weaviate, nothing to set up.

It understands your domain. You can drop a Markdown file at config/glancer/llm_context.glancer.md and teach Glancer your business rules — what status values mean, how revenue is calculated, which columns to ignore:

- `orders.status` values: "pending" | "paid" | "shipped" | "refunded"
- Monthly revenue = SUM(orders.total) WHERE status = "paid"
- When asked about "churn", use the `churned_at` column on `subscriptions`

The UI has some nice touches. Results render in a table with one-click CSV export. Charts are auto-generated where they make sense. Queries run in a background thread, so the main app thread stays free. You can type @table_name to pin a specific table to your question, edit the generated SQL inline and re-run it, and even dictate questions via microphone. There is also a schema viewer at /glancer/db-schema.

Blazer integration. If you already have Blazer installed, Glancer shows a button to open the generated SQL there directly.

The story behind it

This started as a study project. I had never published a gem before and wanted to understand the full process, from engine internals to gemspec to CI. I shared it with a few colleagues and they actually found it useful, so I decided to put it out there for the community.

It is version 1.0.0.

If you try it, I would love to hear what you think, about the code or the idea itself. Issues and pull requests are open.

If you find it useful, a star on the repo goes a long way in helping more people discover it.

GitHub: https://github.com/ErnaneJ/glancer
RubyGems: https://rubygems.org/gems/glancer