Glancer

Requirements

Glancer is built on top of ruby_llm, a provider-agnostic LLM client for Ruby. All LLM calls (query generation, humanized responses, embeddings, and optional question enrichment) go through ruby_llm, so any model it supports works with Glancer.

Installation

2. Run the install generator

rails generate glancer:install

This creates:

• config/initializers/glancer.rb — your main configuration file
• config/glancer/llm_context.glancer.md — optional domain context written in Markdown
• Mounts the engine at /glancer in config/routes.rb

Usage from Ruby

You can call Glancer's internals directly from the Rails console or your own code:

```ruby

Configuration

Edit config/initializers/glancer.rb. Minimal working setup:

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  # allow indexing db/schema.rb
end

Full configuration reference

Chat interface

Visit /glancer in your browser.

• Async processing — messages are handled in a background thread; the UI polls for completion so you can open a new chat while another query runs.
• Step labels — the interface shows what the pipeline is doing: enriching, retrieving context, generating code, validating, executing, preparing response.
• @mention autocomplete — type @table_name to pin a specific table to your question; it renders as a chip linked to the schema viewer.
• Dual query modes — generated SQL or Ruby is syntax-highlighted in the response.
• Inline code editing — modify the generated query and re-run it without asking a new question. Edited versions show a badge.
• Results table — with one-click CSV export (client-side, no backend endpoint).
• Charts — bar, line, doughnut, and scatter charts are auto-generated from the result set where meaningful.
• Fullscreen charts — expand any chart to a fullscreen dialog for detailed inspection.
• Blazer integration — open the SQL in Blazer pre-filled, if the gem is installed.
• Audio input — click the microphone to dictate your question.
• Multi-language — ask in any language; the LLM responds in the same language.
• Custom instructions — set persistent system instructions at /glancer/settings.
• Schema viewer — browse all indexed tables and columns at /glancer/db-schema.
• Message details panel — shows generated code, edit history, execution audit, sources used, and enriched question.

Validate table references against the indexed schema

Glancer::Workflow::SQLValidator.validate_tables_exist!("SELECT * FROM orders JOIN unknown_table") ```

Run the full pipeline

result = Glancer::Workflow.run(chat.id, "Which products have never been ordered?")

5. Start asking questions

http://localhost:3000/glancer

Retrieve relevant chunks for a question (without running the full pipeline)

chunks = Glancer::Retriever.search("monthly revenue by region")