GPT

Use OpenAI, Anthropic, or Gemini models inside Visual Studio Code

🚀 Quick Start (30 seconds)

Prerequisite: an API key from your provider (OpenAI / Anthropic / Gemini).

1. Run GPT: Setup
2. Select any text in the editor
3. Press Alt + Shift + I (Ask GPT)

No key set? Ask GPT will prompt you to set one and automatically retry once.

Contents

• How it works
• How to use
• 🔑 Setup & API keys
• 🧠 Providers & models
• 🧩 Project instructions (.gpt-instruction)
• 🧰 Commands
• 🛠️ Troubleshooting
• 💙 Support
• License

How it works

• You select text (or a full file) → GPT sends that content to the currently selected model/provider and returns a response.
• Responses are shown either in a new file or by replacing the selection (configurable).
• API keys are stored securely in VS Code SecretStorage when available.

Privacy & security

• API keys: stored in VS Code SecretStorage when available; otherwise stored for the current session only.
• Logging: debug logs intentionally avoid secrets and prompt contents.
• Chat history: kept in-memory during the session; you can export it manually.

How to use

Ask GPT (selection) — recommended default

1. Select text
2. Run Ask GPT (Alt + Shift + I)

Ask GPT with File (whole file)

Use Ask GPT with File when you need full-file context (refactors, audits, “explain this file”).

🔑 Setup & API keys

GPT: Setup

One onboarding flow:

1. Choose a provider (OpenAI / Anthropic / Gemini)
2. Set an API key if missing
3. Optionally pick a model

GPT: Set API Key (fast path)

Sets/updates the API key for the currently selected provider.

GPT: Manage API Keys

Manage keys across providers (view status, set/update, remove).

Legacy OpenAI keys previously stored in globalState are migrated automatically when SecretStorage is available.

🧠 Providers & models

• Default provider: OpenAI
• Switch provider: GPT: Change Provider
• Change model: GPT: Change Model

Model list behavior

• Pick from a built-in list or use Custom model id…
• Refresh model list (online) appears only after an API key is configured for the active provider

🧩 Project instructions (.gpt-instruction)

Place a .gpt-instruction file in your workspace to automatically prefix each query with project-specific guidance.

• Multi-root: resolves per workspace folder based on the active document.

• Lookup modes:

  • workspaceRoot (default): reads <workspaceFolder>/.gpt-instruction
  • nearestParent: closest parent .gpt-instruction wins

• Nearest-parent behavior:

  • An empty .gpt-instruction suppresses parent instructions.
  • This mode can be more expensive in very large repos (recursive watcher).

• Size limits:

  • Content beyond the configured max is truncated with a warning.
  • In remote/virtual workspaces, very large files may be ignored for safety.

🧰 Commands

Core

• GPT: Setup — onboarding
• Ask GPT — run on selection (Alt + Shift + I)
• Ask GPT with File — run on entire file
• GPT: Set API Key — set key for active provider
• GPT: Manage API Keys — manage keys across providers

Provider & model

• GPT: Change Provider
• GPT: Change Model
• GPT: Change Token Limit

Conversation & output

• GPT: Change Output Mode — Replace Selection vs New File
• GPT: Change Context Mode — No Context / Last N / Full
• GPT: Set Context Length
• GPT: Export Chat History
• GPT: Show Chat History
• GPT: Clear Chat History

Debug

• GPT: Change Debug Mode — show/hide the GPT Debug output channel

🛠️ Troubleshooting

Invalid or missing API key

• Run GPT: Setup (recommended) or GPT: Set API Key
• Confirm provider via GPT: Change Provider

“Model not available”

• Run GPT: Change Model
• If a key is configured, use Refresh model list (online) or choose Custom model id…

Rate limits / quota

• Retry later or check your provider plan/quota.
• OpenAI usage dashboard: https://platform.openai.com/account/usage

Debugging

• Toggle GPT: Change Debug Mode
• Check the GPT Debug output channel (no secrets or prompts are logged)

Support

If this extension saves you time, you can support ongoing maintenance:

• ☕ Buy Me a Coffee: https://buymeacoffee.com/silasnevstad
• ⭐ GitHub Sponsors: https://github.com/sponsors/silasnevstad

License

MIT License — see LICENSE