Universal Provider Support
Connect to built-in providers (OpenAI, ChatGPT/Codex OAuth, xAI Grok OAuth, OpenRouter), custom Anthropic and Google Gemini adapters, and local OpenAI-compatible endpoints (Ollama, llama.cpp).
Universal Provider Support
Connect to built-in providers (OpenAI, ChatGPT/Codex OAuth, xAI Grok OAuth, OpenRouter), custom Anthropic and Google Gemini adapters, and local OpenAI-compatible endpoints (Ollama, llama.cpp).
Execution Safety & Sandboxing
Sandboxed shell execution with configurable read/write boundaries (standard and strict), interactive diff previews, smart auto-approvals, and run budgets.
Five Operating Modes
Standard, Plan, Lite, Mentor, and Orchestrator modes adapted to research, planning, full development, collaborative second-opinions, and delegated orchestration.
Multi-Agent Swarms
Deploy specialized subagents (Explorer, Worker, Mentor, Librarian, and Reviewer) across tiered model profiles (smart, balanced, cheap, chore) in foreground or background with Ctrl+G task management.
Scriptable Tool Workflows
Use run_code to coordinate tool calls in a sandboxed JavaScript runtime with nested approvals, structured results, and explicit return contracts.
Native Remote Development
Connect directly to remote servers over SSH with local SSH agent authentication and run full or lightweight agent sessions remotely.
Private Web Gateway
Launch term2 serve to expose an authenticated, audited Unix socket or TLS gateway for programmatic control and external clients.