AI Autocomplete
leoyishou114 downloadsInline AI writing completion with ghost text. Powered by Groq. Tab to accept.
- Overview
- Scorecard
- Updates6
AI inline writing completion for Obsidian, powered by OpenAI-compatible APIs. The default setup routes through OpenRouter to the Groq provider for fast inference.
Type naturally and get ghost text suggestions that appear inline. Press Tab to accept, Esc to dismiss.
Features
- Ghost text completion — transparent suggestions appear at your cursor, like GitHub Copilot
- Context-aware — reads text before and after cursor for coherent continuations
- Insight-oriented — can surface sharper questions, hidden assumptions, analogies, and reframes for personal knowledge notes
- Fast — defaults to OpenRouter's Groq provider with throughput-prioritized routing
- Bilingual — automatically detects and continues in Chinese or English
- Lightweight — 6KB plugin, no dependencies
Usage
- Install the plugin
- Go to Settings → AI Autocomplete → enter your OpenRouter API key
- Start writing — suggestions appear after a brief pause
| Key | Action |
|---|---|
| Tab | Accept suggestion |
| Esc | Dismiss suggestion |
| Keep typing | Suggestion auto-dismisses |
Settings
| Setting | Default | Description |
|---|---|---|
| API base URL | https://openrouter.ai/api/v1/chat/completions |
Any OpenAI-compatible chat completions endpoint |
| Model | openai/gpt-oss-120b:nitro |
Smart default via OpenRouter's Groq provider |
| System prompt | Built-in heuristic prompt | Editable prompt that controls ghost text style and insight behavior |
| Reasoning effort | minimal |
Keeps reasoning models fast enough for inline autocomplete |
| Hide reasoning | On | Excludes reasoning tokens from suggestion text |
| Provider | groq |
Forces OpenRouter's Groq provider |
| Provider sort | throughput |
Prioritizes high token throughput |
| Allow fallbacks | Off | Keeps requests on the selected provider |
| Trigger delay | 800ms | How long to wait after typing before fetching a suggestion |
| Enabled | On | Toggle via settings or command palette |
How it works
The plugin uses CodeMirror 6 extensions to render transparent "ghost text" at the cursor position. When you pause typing, it sends the surrounding context (up to 2000 chars before + 500 chars after cursor) to the configured API and displays the completion as inline ghost text.
License
MIT
83%
HealthExcellent
ReviewSatisfactory
About
Show inline ghost-text suggestions as you type, offering transparent continuations and insight-driven prompts that read text before and after the cursor. Generate fast, bilingual (English/Chinese) completions via OpenAI-compatible APIs and accept suggestions with Tab or dismiss with Esc.Details
Current version
1.0.5
Last updated
3 weeks ago
Created
3 months ago
Updates
6 releases
Downloads
114
Compatible with
Obsidian 0.15.0+
Platforms
Desktop, Mobile
License
MIT