Ollama Sidebar

A browser sidebar chat extension that connects to a self-hosted Ollama instance. Built for Orion (WebKit) with full Chrome support. No cloud, no accounts, no bundler — loads unpacked directly from disk.

Requirements

Ollama running locally or on a remote host
Chrome 114+ or Orion (beta Chrome extension support)
At least one model pulled in Ollama
OLLAMA_ORIGINS=* set on the Ollama instance if proxied through Nginx or another reverse proxy

Browser Compatibility

Browser	Status	Notes
Orion	✅ Main supported browser	File picker button does not open a picker — drag and drop files onto the chat instead. Under investigation.
Chrome	✅ Fully working	All features including file picker. `sidebar_action` manifest warning is expected and harmless.
Firefox	🔜 Not yet supported	Possible future support.

Installation

Clone or download this repository
Open the icons/ folder — if the PNG files are missing, open generate_icons.html in a browser and save the three icons into the icons/ folder
Chrome: go to chrome://extensions, enable Developer Mode, click Load unpacked, select the project folder
Orion: go to Tools → Extensions → Install from Disk, select the project folder
Click the extension icon in the toolbar to open the sidebar

First-time setup

Click the ⚙ Settings button in the sidebar header
Set your Ollama Base URL (default: http://localhost:11434)
Click Test Connection — connected models will appear in the dropdowns
Set a Default Chat Model
Click Save Settings

If Ollama is behind a reverse proxy with bearer token auth, enter the token in the Bearer Token field.

Features

Chat

Streaming responses with stop, regenerate, and edit
Thinking/reasoning block support (collapsed by default)
Full markdown rendering with syntax-highlighted code blocks
Token metrics per response (tokens, tok/s, duration)

Context

Add the current page as context with the 🌐 button — uses a Readability-style algorithm to extract main content, strips navigation and noise
Drag and drop files onto the chat to attach (.txt, .md, .csv, .pdf, images)
PDF text extraction via PDF.js
Image support for vision-capable models (llava, qwen-vl, etc.)
Keyword and semantic RAG — relevant chunks are selected before being sent, scaled to your model's context window

Sessions

Multiple named chat sessions with create, delete, search, and export as Markdown
Auto-generated session names from the first message
Session history persisted in browser storage

Personas

8 built-in personas (Assistant, Code Reviewer, Study Buddy, Writing Editor, Rubber Duck, Devil's Advocate, Technical Writer, Be Brief/BLUF)
Create and save custom personas
Each session tracks its active persona's system prompt

Prompt Templates

10 built-in templates triggered with / in the input
Create and save custom templates

Models

Per-session model selector
Pull new models from the Settings page without leaving the browser
Multi-model comparison mode — send the same prompt to two models side by side
Semantic RAG via Ollama /api/embed endpoint (configure embedding model in Settings)

Settings

Dark/light theme
Compact message density
Auto-scroll during generation
Response language lock
Model parameters (temperature, top-p, top-k, repeat penalty, seed, context length, max tokens)
Default system prompt
Debug panel with live log, system info, and connection status

Keyboard shortcuts

Shortcut	Action
`Enter`	Send message
`Shift+Enter`	New line
`Ctrl+K`	New chat
`Ctrl+L`	Clear conversation (press twice to confirm)
`Ctrl+/`	Open templates
`Ctrl+,`	Open settings
`Ctrl+?`	Show all shortcuts
`Esc`	Close panels and overlays

Project structure

├── background.js          # Service worker — message routing, stream relay
├── content.js             # Content script — page text extraction
├── sidepanel.html         # Sidebar UI
├── sidepanel.js           # Sidebar logic
├── sidepanel.css          # Sidebar styles
├── settings.html          # Options page
├── settings.js            # Options page logic
├── settings.css           # Options page styles
├── manifest.json          # Extension manifest (MV3)
├── marked.min.js          # Markdown rendering
├── highlight.min.js       # Syntax highlighting
├── hljs-github-dark.min.css  # Highlight.js theme
├── pdf.min.js             # PDF.js v3 — PDF text extraction
├── pdf.worker.min.js      # PDF.js worker
├── pdf-init.js            # PDF.js worker initialisation
└── icons/
    ├── icon16.png
    ├── icon48.png
    └── icon128.png

Notes

File upload: Chrome supports the native file picker via the attach button. Orion does not currently open the file picker — drag and drop files directly onto the chat area instead. This is under investigation.
Semantic RAG: Requires an embedding model pulled in Ollama (recommended: nomic-embed-text-v2-moe). Configure it under Settings → Models → Embedding Model. If not configured, keyword matching is used.
Multi-model comparison: Results are not saved to session history.
Debug panel: Available in Settings → Debug (collapsed by default). Logs auto-clear after 30 minutes.
Ollama CORS: If requests are blocked, ensure OLLAMA_ORIGINS=* is set in your Ollama environment or your reverse proxy passes the correct CORS headers.

License

MIT

5.6 KiB Raw Blame History