Ghostwrite

AI that assists. You that writes.
Inline code completion powered by Claude, GPT-4o, and Gemini—for developers who want LLM intelligence without giving up the keyboard.

Quick Start • Features • Configuration • How It Works • Privacy & Cost

Why Ghostwrite?

AI coding tools have a philosophy problem.

Most assume you want to write less code. They optimize for delegation—agents that scaffold entire features, chat interfaces that generate files wholesale, autonomous systems that treat your codebase as a problem to be solved without you.

That's useful sometimes. But it's not the only way to work.

Ghostwrite is for developers who still want to write code.

If you find value in staying close to your logic. If you think better when your hands are on the keyboard. If you've noticed that "just let the AI do it" often means "now debug what the AI did"—Ghostwrite was built for how you work.

What makes it different

Inline only. No chat panels. No autonomous agents. No context-switching. Completions appear where you're typing, when you're ready for them.

Bidirectional context. Most tools only see what's above your cursor. Ghostwrite analyzes code before and after—so completions fit the shape of what you're building, not just what you've built.

function processUser(user: User) {
  // Ghostwrite sees this context above ↑
  |  // ← Your cursor is here
  // AND this context below ↓
  return enrichedUser;
}

Smart boundaries. Completions stop at logical endpoints. No runaway generation that overshoots your intent and leaves you cleaning up.

Your choice of model. Claude, GPT-4o, or Gemini. Use what works for your context and budget. Switch anytime.

Full control. Configure token limits, context windows, trigger delays, and debounce timing. The tool adapts to you, not the reverse.

Who this is for

Engineers who want assistance, not automation
Developers who've felt alienated by "vibe coding" culture
Teams that value code comprehension alongside velocity
Anyone who's tired of tools that assume they want to write less code

Who this is not for

If you want AI to write entire features autonomously
If you prefer chat-based code generation
If you're optimizing purely for lines-of-code-per-hour

We'd rather be honest: those workflows are valid, but they're not what Ghostwrite is designed for.

Technical differentiators

Capability	Ghostwrite	Agentic Tools
Philosophy	You write the code	AI writes for you
Interface	Inline completions	Chat panels, autonomous agents
Context Analysis	Bidirectional (before + after cursor)	Varies
Completion Boundaries	Smart (stops at logical endpoints)	Often overshoots
Model	Your choice (Claude, GPT-4o, Gemini)	Usually locked
Configuration	Full control (tokens, context, timing)	Limited
Secret Detection	Built-in (blocks credentials in context)	Rarely included

Quick Start

1. Choose Your Provider

Ghostwrite supports three AI providers. Pick one and get an API key:

Provider	Sign Up	Environment Variable
Anthropic (default)	console.anthropic.com	`ANTHROPIC_AUTH_TOKEN`
OpenAI	platform.openai.com	`OPENAI_API_KEY`
Google Gemini	aistudio.google.com	`GEMINI_API_KEY`

2. Set Environment Variables

# Add to your shell profile (~/.zshrc, ~/.bashrc, etc.)

# Option A: Anthropic (default)
export ANTHROPIC_AUTH_TOKEN="sk-ant-..."

# Option B: OpenAI
export GHOSTWRITE_PROVIDER="openai"
export OPENAI_API_KEY="sk-..."

# Option C: Gemini
export GHOSTWRITE_PROVIDER="gemini"
export GEMINI_API_KEY="..."

Restart your terminal or run source ~/.zshrc.

3. Install Ghostwrite

Clone and install locally:

git clone https://github.com/ghostwrite-ai/ghostwrite.git
cd ghostwrite
npm install
npm run package

Then install the generated .vsix file: Extensions > ... > Install from VSIX

4. Start Coding

Open any file and start typing. Completions appear automatically after a brief pause.

That's it. Press Tab to accept, Escape to dismiss.

Features

Bidirectional Context

Ghostwrite analyzes code before and after your cursor position. This means suggestions understand:

The function you're implementing
The return type you need to satisfy
Variables already declared below
Patterns used throughout the file

Smart Boundary Detection

Completions stop at natural code boundaries:

End of statements (;, ,)
Closing braces (}, ), ])
Logical break points

No more accepting a suggestion only to delete half of it.

Streaming Completions

See suggestions as they generate. Accept early if the first part is what you need, or wait for the complete suggestion.

Built-in Secret Detection

Ghostwrite automatically blocks completions when it detects potential secrets in your context:

API keys (AWS, GitHub, Stripe, Anthropic, OpenAI)
Passwords and auth tokens
Private keys
Database connection strings

The status bar shows a shield icon when secrets are detected.

Intelligent Request Management

Debouncing: Waits for you to pause typing before requesting
Caching: Reuses valid completions as you type forward
Cancellation: Aborts in-flight requests when you keep typing
Retry Logic: Automatically retries on transient failures

Configuration

Access settings via Preferences > Settings > Ghostwrite or add to settings.json:

{
  "ghostwrite.enabled": true,
  "ghostwrite.maxTokens": 256,
  "ghostwrite.debounceMs": 300,
  "ghostwrite.contextLines": 100,
  "ghostwrite.secretDetection": true,
  "ghostwrite.timeoutMs": 30000
}

Settings Reference

Setting	Default	Range	Description
`enabled`	`true`	—	Enable/disable completions globally
`maxTokens`	`256`	32–1024	Maximum tokens per completion
`debounceMs`	`300`	50–2000	Delay before triggering (ms)
`contextLines`	`100`	10–500	Lines of context in each direction
`secretDetection`	`true`	—	Block when secrets detected
`timeoutMs`	`30000`	5000–120000	Request timeout (ms)

Tuning Recommendations

For faster suggestions: Lower debounceMs to 150-200 (increases API calls)

For longer completions: Increase maxTokens to 512 (increases cost)

For large files: Increase contextLines to 200 (increases cost, improves context)

For cost control: Lower maxTokens to 128 and contextLines to 50

Commands

Command	Description	Default Keybinding
`Ghostwrite: Toggle On/Off`	Enable/disable completions	—

Access via Command Palette (Cmd/Ctrl+Shift+P).

Status Bar

The status bar indicator shows Ghostwrite's current state:

Icon	State	Meaning
👻	Ready	Waiting for input
⏳	Loading	Generating completion
⊘	Disabled	Extension is off
⚠️	Error	Auth, quota, or network issue
🛡️	Blocked	Secrets detected in context

Click the status bar to toggle Ghostwrite on/off.

How It Works

┌─────────────────────────────────────────────────────────────────┐
│                         Your Editor                              │
├─────────────────────────────────────────────────────────────────┤
│  ... code above cursor (up to 100 lines) ...                    │
│                                                                  │
│  function example() {                                            │
│    const data = |  ← cursor position                            │
│    return processed;                                             │
│  }                                                               │
│                                                                  │
│  ... code below cursor (up to 100 lines) ...                    │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                    Ghostwrite Extension                          │
├─────────────────────────────────────────────────────────────────┤
│  1. Extract bidirectional context                                │
│  2. Scan for secrets (block if found)                           │
│  3. Build prompt with language + filename                       │
│  4. Stream completion from your chosen provider                 │
│  5. Stop at logical boundary                                    │
│  6. Present as inline suggestion                                │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                    AI Provider (your choice)                     │
├─────────────────────────────────────────────────────────────────┤
│  • Claude Opus 4.5 (Anthropic)                                  │
│  • GPT-4o (OpenAI)                                              │
│  • Gemini 1.5 Pro (Google)                                      │
└─────────────────────────────────────────────────────────────────┘

Privacy & Cost

What Data Is Sent

When generating completions, Ghostwrite sends to your chosen provider's API:

Code before your cursor (configurable, default 100 lines)
Code after your cursor (configurable, default 100 lines)
File language and name
Your system username (for rate limiting, Anthropic only)

No code is stored locally beyond the current session. See your provider's privacy policy for server-side handling:

Secret Detection

Enabled by default, Ghostwrite scans context for:

AWS access keys and secrets
GitHub tokens (PAT, OAuth)
Stripe API keys
Private keys (RSA, DSA, EC)
Database connection strings
Generic API keys and passwords

When detected, completions are blocked and the status bar shows 🛡️.

API Costs

All providers charge based on token usage. Costs vary by provider:

Provider	Model	Approx. Cost per 1K tokens
Anthropic	Claude Opus 4.5	~$0.015 input, ~$0.075 output
OpenAI	GPT-4o	~$0.005 input, ~$0.015 output
Google	Gemini 1.5 Pro	~$0.00125 input, ~$0.005 output

Monitor usage: Check your provider's dashboard.

Control costs: Reduce maxTokens and contextLines settings.

Environment Variables

Provider Selection

Variable	Default	Description
`GHOSTWRITE_PROVIDER`	`anthropic`	Provider to use: `anthropic`, `openai`, or `gemini`

Anthropic

Variable	Required	Description
`ANTHROPIC_AUTH_TOKEN`	Yes*	Anthropic API key
`ANTHROPIC_BASE_URL`	No	Custom API endpoint (proxies, enterprise)
`ANTHROPIC_MODEL`	No	Model override (default: `claude-opus-4-5-20251101`)

OpenAI

Variable	Required	Description
`OPENAI_API_KEY`	Yes*	OpenAI API key
`OPENAI_BASE_URL`	No	Custom API endpoint (Azure, proxies)
`OPENAI_ORGANIZATION`	No	Organization ID for multi-org accounts
`OPENAI_MODEL`	No	Model override (default: `gpt-4o`)

Gemini

Variable	Required	Description
`GEMINI_API_KEY`	Yes*	Google AI API key
`GEMINI_MODEL`	No	Model override (default: `gemini-1.5-pro`)

*Required only when using that provider.

Troubleshooting

"Invalid API Key" Error

Verify your API key is set for your provider:
- Anthropic: echo $ANTHROPIC_AUTH_TOKEN
- OpenAI: echo $OPENAI_API_KEY
- Gemini: echo $GEMINI_API_KEY
Check the key is valid at your provider's dashboard
Restart VS Code after setting the variable

"Unknown Provider" Error

Ensure GHOSTWRITE_PROVIDER is set to one of: anthropic, openai, or gemini (case-insensitive).

No Completions Appearing

Check status bar — is Ghostwrite enabled?
Type some code and pause (completions need a trigger delay)
Check Output panel (View > Output > Ghostwrite) for errors

Completions Are Slow

Reduce contextLines to 50
Reduce maxTokens to 128
Check your network connection

"Secrets Detected" Blocking

If you're working with test credentials or examples:

Disable temporarily: "ghostwrite.secretDetection": false
Move secrets to environment variables
Use .env.example with placeholder values

Architecture

src/
├── extension.ts          # VS Code extension entry point
├── completionProvider.ts # Inline completion logic
├── contextBuilder.ts     # Bidirectional context extraction
├── secretDetector.ts     # Credential scanning
├── errorHandler.ts       # Error classification & retry logic
├── requestManager.ts     # Debouncing & request lifecycle
├── statusBar.ts          # Status indicator management
├── types.ts              # TypeScript interfaces
└── providers/
    ├── types.ts          # Provider interfaces & config types
    ├── anthropicAdapter.ts # Anthropic SDK adapter
    ├── openaiAdapter.ts    # OpenAI SDK adapter
    ├── geminiAdapter.ts    # Gemini SDK adapter
    ├── configParser.ts     # Environment config parsing
    └── providerFactory.ts  # Provider instantiation

Built with:

TypeScript — Full type safety, zero any, no type assertions
neverthrow — Type-safe error handling with Result<T, E>
Adapter pattern — Provider-agnostic streaming interface
Functional patterns — Immutable data, pure functions, discriminated unions

Development

# Clone the repository
git clone https://github.com/ghostwrite-ai/ghostwrite.git
cd ghostwrite

# Install dependencies
npm install

# Compile TypeScript
npm run compile

# Watch mode for development
npm run watch

# Package for distribution
npm run package

Running Locally

Open the project in VS Code
Press F5 to launch Extension Development Host
Test completions in the new VS Code window

Contributing

Contributions are welcome! Please:

Fork the repository
Create a feature branch: git checkout -b feature/amazing-feature
Make your changes
Run npm run compile to verify
Commit: git commit -m 'Add amazing feature'
Push: git push origin feature/amazing-feature
Open a Pull Request

License

MIT — Use freely, attribution appreciated.

Built for developers who still want to write code.
Report Bug • Request Feature

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
.vscode		.vscode
src		src
.gitignore		.gitignore
.vscodeignore		.vscodeignore
CHANGELOG.md		CHANGELOG.md
LICENSE		LICENSE
README.md		README.md
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json

License

ryannono/Ghostwrite

Folders and files

Latest commit

History

Repository files navigation

Ghostwrite

Why Ghostwrite?

What makes it different

Who this is for

Who this is not for

Technical differentiators

Quick Start

1. Choose Your Provider

2. Set Environment Variables

3. Install Ghostwrite

4. Start Coding

Features

Bidirectional Context

Smart Boundary Detection

Streaming Completions

Built-in Secret Detection

Intelligent Request Management

Configuration

Settings Reference

Tuning Recommendations

Commands

Status Bar

How It Works

Privacy & Cost

What Data Is Sent

Secret Detection

API Costs

Environment Variables

Provider Selection

Anthropic

OpenAI

Gemini

Troubleshooting

"Invalid API Key" Error

"Unknown Provider" Error

No Completions Appearing

Completions Are Slow

"Secrets Detected" Blocking

Architecture

Development

Running Locally

Contributing

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Languages

Packages