Skip to main content
Glass provides a comprehensive suite of AI capabilities powered by multiple language model providers. The AI features are deeply integrated into the editor to enhance your coding workflow.

​
Core AI Features

Glass offers three primary AI-powered features:

Agent Panel

Interactive AI agents for code generation, refactoring, and complex tasks

Inline Assistant

Context-aware AI assistance directly in your editor

Code Completion

Real-time AI-powered code suggestions as you type

​
Supported AI Providers

Glass supports multiple language model providers, allowing you to choose the best option for your workflow:
  • Anthropic Claude - Claude 4.6 Sonnet, Opus, and other Claude models
  • OpenAI - GPT-4, GPT-5, o1, o3, and other OpenAI models
  • OpenAI-compatible APIs - Any OpenAI-compatible endpoint
  • Ollama - Local models for privacy and offline usage
  • Google Gemini - Gemini models via Agent Servers
  • Custom providers - Extend with your own model providers
Most AI features require authentication with at least one provider. See the Provider Configuration guide for setup instructions.

​
Agent Types

Glass supports multiple agent implementations:

​
Native Agent (Zed Agent)

The built-in agent with full access to Glass features:
  • Tool support - Can edit files, run terminal commands, search code
  • Context awareness - Accesses project structure, rules, and prompts
  • Multi-model support - Works with any configured provider
  • Streaming responses - Real-time response generation

​
Text Thread Agent

A lightweight agent for conversational interactions:
  • Simple conversations - No tool execution
  • Fast responses - Optimized for quick interactions
  • Context attachment - Manual file and context inclusion

​
External Agents

Support for third-party agent implementations:
  • Claude Code (via Agent Server) - Official Anthropic agent
  • Gemini CLI - Google’s agent implementation
  • Codex - OpenAI’s code-focused agent
  • Custom agents - Via Agent Client Protocol (ACP)

​
Key Capabilities

​
Context Management

1

Project Context

Agents automatically access project structure, worktree information, and rules files (.zed.md, .rules, etc.)
2

Related Files

Semantic search finds relevant code files based on your current task
3

Prompt Library

Reusable prompts stored in ~/.config/zed/prompts/ for common tasks
4

Context Servers

Model Context Protocol (MCP) servers provide external context sources

​
Tool Execution

Native agents can execute tools with configurable permissions:
  • File Operations - Read, write, edit files with diff preview
  • Terminal Commands - Execute shell commands with approval
  • Code Search - Search across project with semantic and text search
  • Git Operations - View diffs, branch management
  • LSP Integration - Go to definition, find references
Tool permissions can be configured in settings. By default, most operations require confirmation.

​
Model Selection

Choose different models for specific tasks: 
{
  "agent": {
    "default_model": {
      "provider": "anthropic",
      "model": "claude-4.6-sonnet"
    },
    "inline_assistant_model": {
      "provider": "anthropic",
      "model": "claude-4.6-sonnet"
    },
    "thread_summary_model": {
      "provider": "openai",
      "model": "gpt-4o-mini"
    }
  }
}

​
AI Configuration

Provider Setup

Configure API keys and endpoints for each provider

Settings

Customize AI behavior, tool permissions, and model parameters

​
Privacy and Data

When using AI features, Glass sends:
  • Your prompts and messages
  • Relevant file contents (based on context selection)
  • Project structure information
  • Code snippets from active files
You control what context is included via context servers and file selections.
For complete privacy, use Ollama or other local providers:
  • No data leaves your machine
  • Works offline
  • Full control over models
See Provider Configuration for Ollama setup.
API keys are stored securely:
  • macOS: Stored in Keychain
  • Linux: Stored in Secret Service (gnome-keyring, KWallet)
  • Windows: Stored in Credential Manager
Keys can also be set via environment variables.

​
Getting Started

1

Configure a Provider

Set up at least one AI provider in Settings
2

Open Agent Panel

Use cmd-shift-a (macOS) or ctrl-shift-a (Linux/Windows)
3

Start a Conversation

Type your question or task and press Enter
4

Review and Apply

Review suggested changes and apply them to your code

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us