Ambiance

Ambiance

Semantic search for documents and code on macOS.

Download for macOS

Version 1.3.0

Ambiance search window

Features

Semantic Search

Find documents by meaning, not just keywords. A trained byte-level transformer encoder maps your query and documents into a shared embedding space for cosine similarity search via HNSW.

Text & Regex Search

Switch between semantic, case-insensitive text, and POSIX extended regex modes with a single click. Find exact phrases or complex patterns across all indexed files.

Per-Directory Tabs

Each directory gets its own tab with an independent search index. Drag to reorder, see the Git branch and chunk count at a glance. Restored on relaunch.

Train Your Own Encoder

Create training profiles for different languages or domains. Point at directories of documents, click train, and get a domain-tuned encoder in minutes on Apple Silicon.

Native PDF Support

PDFs are extracted using the built-in system framework — no package managers, no command-line utilities, no external tools. Double-click a result to open the PDF at the matching page.

Pinned Results

Pin important results to the top of the list. Pinned results persist across searches and app restarts, making it easy to curate reference material per directory.

RAG Prompt Builder

Format search results as a context-enriched prompt for ChatGPT or Claude. Choose pinned, top 5, or all visible results, click Copy Prompt, and paste into your LLM.

Rope Integration

Double-click a source code result to open it in Rope at the exact line via the rope:// URL scheme. PDFs open in Preview at the matching page.

Smart File Detection

Uses Apple's UTType system to classify files. Source code, text, PDFs, and markup are indexed automatically. Binary files, images, and compiled objects are skipped.

How It Works

  1. Train an encoder — Point the trainer at your documents to learn domain-specific byte patterns. Swedish documents? English research papers? Train a profile for each.
  2. Open a directory — Ambiance indexes all text files, source code, and PDFs into overlapping chunks stored in a local HNSW index.
  3. Search — Type a natural language query for semantic search, an exact phrase for text search, or a pattern for regex. Results show file, line, score, and snippet.
  4. Use results — Double-click to open in Rope or Preview. Pin important findings. Copy results as a RAG prompt for your favourite LLM.

Keyboard Shortcuts

⌘⇧O
Open Directory
⌘W
Close Tab
⌘⇧R
Re-index
⌘I
File Scope
⌘⇧C
Copy Prompt
⌘,
Settings
⌘?
Help
Return
Search

Requirements

  • macOS 14 (Sonoma) or later
  • Apple Silicon recommended for training
  • Rope (optional, for source code navigation)