Features

Multi‑Provider Architecture

Arctic supports 130+ AI providers in one unified interface:

Coding Plans (Subscription-Based)

Codex (ChatGPT) - Latest GPT-5 models with coding optimizations
Gemini CLI - Google's Gemini models with Code Assist
Antigravity - Gemini 3 Pro and Claude models
GitHub Copilot - Including Enterprise support
Claude Code - Anthropic's coding-focused plans
Z.AI - Specialized coding models
Kimi for Coding - Long-context coding assistance
Amp Code - GPT-5 series models
Qwen Code - Alibaba's coding models (1M context)
MiniMax - Coding plan support

API Providers

OpenAI - GPT-4o, o1, o3-mini and more
Anthropic - Claude Sonnet, Opus, Haiku
Google - Gemini Pro, Flash, Thinking
Amazon Bedrock - Nova, Claude, Llama models
Azure - OpenAI and Cognitive Services
Perplexity - Sonar models
OpenRouter - 200+ models aggregated
Ollama - Local model support
Groq - Ultra-fast inference
Together AI - Open source models
DeepSeek - Reasoning models
Cerebras - High-performance inference
And 100+ more...

Multiple Accounts Per Provider

Connect multiple accounts for the same provider (e.g., personal + work):

Separate usage tracking
Easy switching between accounts
Independent rate limits

Mid-Conversation Model Switching

Switch models during a conversation without losing context:

Press Ctrl+X M in TUI
Use --model flag in CLI
Seamless context transfer

Usage Tracking & Analytics

Real-Time Usage Limits

Track usage across all providers in one view
Rolling windows (5-hour, daily, weekly limits)
Credit balance monitoring
Rate limit warnings

Comprehensive Statistics

View detailed analytics with arctic stats:

Activity Heatmap

GitHub-style contribution graph
Daily coding activity visualization
Streak tracking
Active days counter

Token Analytics

Input/output token breakdown
Cache read/write tracking
Cost per token type
"War and Peace" comparison (730k tokens)

Cost Analysis

Total spend across all models
Cost per model breakdown
Average cost per session/day
Visual cost bars

Usage Patterns

Favorite models
Peak coding hours
Session statistics
Model usage distribution

Skills System

Skills are reusable prompt templates that can be discovered and loaded on-demand:

Store specialized instructions for specific tasks
Searchable by name and description
Tool restrictions per skill
Auto-discovery by AI
Share across projects

Example skills:

Code reviews with security focus
Git release workflows
Documentation generation
Test creation
Security audits

See Skills for details.

Agents & Commands

Agents

Custom assistants with specialized behavior:

Define conversation style and persona
Set tool permissions
Configure model preferences
Support for primary and subagent modes
Nested agent directories

Commands

Templated prompts with variables:

Quick shortcuts for common tasks
Variable substitution
Model and agent overrides
Slash command integration
Import from Claude Code

Modes

Deprecated in favor of agents, but still supported for backward compatibility.

Advanced Tool System

Built-in Tools

File Operations: read, write, edit, glob, grep, list
Shell: bash with command pattern permissions
Web: webfetch, websearch, codesearch (with EXA)
Orchestration: task (spawn sub-agents)
Organization: todowrite, todoread (per-session todos)
Performance: batch (parallel tool execution)
Skills: skill_find, skill_use (dynamic skill loading)

Custom Tools

Add your own tools via:

Plugins: TypeScript/JavaScript tools in .arctic/plugin/
Tool Directory: Simple tool definitions in .arctic/tool/
MCP Servers: External tool servers via Model Context Protocol

Permission System

Fine-grained control over tool usage:

allow, ask, or deny per tool
Pattern matching for bash commands (e.g., git *)
Per-agent permission overrides
Session-level permission memory
External directory restrictions

Model Context Protocol (MCP)

Connect to external tool servers:

Remote MCP Servers

HTTP/HTTPS endpoints
Header-based authentication
OAuth 2.0 support with dynamic registration
Custom timeout configuration

Local MCP Servers

Stdio-based communication
Environment variable injection
Command-line arguments
Process management

Auto-Import

Automatically import MCP servers from Claude Code:

Project-specific MCPs from .mcp.json
Directory-specific MCPs from ~/.claude.json
Global MCPs from root config
One-time import prompts

TUI (Terminal User Interface)

Keyboard-first workflow optimized for speed:

Session list with search
Model picker (Ctrl+X M)
Agent picker (Ctrl+X A)
Command palette (Ctrl+P)
Theme picker

Editing

Multi-line input with syntax highlighting
Copy/paste with system clipboard
Selection tools
External editor integration (Ctrl+X E)

Visualization

Streaming responses with thinking display
Tool execution feedback
Usage meters
Activity heatmaps
Cost tracking

Slash Commands

Built-in shortcuts:

/models, /agents, /session, /new
/status, /usage, /stats
/mcp, /theme, /help
/undo, /redo, /compact
/rename, /copy, /export
/timeline, /thinking, /benchmark
Custom commands from .arctic/command/

Snapshots & Version Control

Git-based snapshots for safe experimentation:

Automatic Snapshots

Created before file modifications
Linked to sessions and messages
Incremental and efficient

Manual Snapshots

Create checkpoints anytime
Name and tag snapshots
Compare snapshots

Diffing & Revert

View changes between snapshots
Revert to any snapshot
Selective file restoration
Benchmark comparisons

Session Management

Session Features

Auto-save conversations
Resume from any point
Fork sessions for experimentation
Export/import sessions
Session timeline view
Compact history

Project Isolation

Sessions scoped to Git worktrees
Automatic project detection
Cross-project session search
Project-specific configuration

Configuration System

Layered Configuration

Configs merge from multiple sources:

Global: ~/.config/arctic/arctic.json
Parent directories: .arctic/arctic.json
Project: <project>/.arctic/arctic.json
Environment variables
CLI flags

Config Features

JSON/JSONC support
TypeScript config files
Hot reload
Config export/backup
Migration from Claude Code/OpenCode

Customization

Default models (primary + small)
Theme selection
Keybindings
Provider enable/disable lists
Permission profiles
Experimental features

Plugin System

Extend Arctic with custom integrations:

Plugin Capabilities

Config Hook: Modify configuration at runtime
Event Hook: Listen to sessions, messages, tools
Auth Hook: Custom authentication flows
Tool Hook: Add custom tools

Plugin Locations

Global: ~/.config/arctic/plugin/
Project: .arctic/plugin/

Use Cases

Custom analytics
External notifications
Model routing logic
Permission extensions
Provider integrations

Claude Code Migration

Seamless migration from Claude Code:

Auto-Import

First launch detects and imports:

Commands: ~/.claude/commands → .arctic/command/claude/
Agents: ~/.claude/agents → .arctic/agent/claude/
MCP Servers: ~/.claude.json → .arctic/arctic.jsonc

Compatibility

Command format compatible
Agent format compatible (with normalization)
MCP configuration compatible
One-time import prompts

Developer Experience

Performance

Fast startup with lazy loading
Efficient caching
Parallel tool execution
Streaming responses

Reliability

Automatic token refresh
Connection retry logic
Error recovery
Graceful degradation

Privacy

Local-first architecture
No data sent to Arctic servers
Optional anonymous telemetry
Disable with arctic telemetry disable

Cross-Platform

macOS: Intel and Apple Silicon
Linux: x64 and ARM64
Windows: x64 support
Clipboard integration (xclip on Linux)

Next Steps

Explore specific features:

Skills - Reusable prompt templates
Agents - Custom assistants
Commands - Templated shortcuts
MCP - External tool servers
Plugins - Custom extensions
Providers - All 130+ providers

On this page