Skip to content
term2

Overview

term2 is an open-source terminal AI assistant and autonomous agent runtime built for high-velocity software engineering and system administration.

Combining an interactive React (Ink) terminal user interface with an application-owned agent run loop, term2 provides fine-grained execution safety, native multi-agent delegation, provider-neutral context compaction, and multi-provider support—giving developers complete agency over their workflow without vendor lock-in or subscription gates.

Core Capabilities

Section titled “Core Capabilities”
  • Universal Provider Support: Native integration with OpenAI (WebSocket Responses and Chat Completions), ChatGPT/Codex OAuth (browser PKCE), xAI Grok (Responses API with encrypted reasoning and OAuth), and OpenRouter, plus custom adapters for Anthropic, Google Gemini, and local OpenAI-compatible endpoints (Ollama, llama.cpp, vLLM).
  • Fine-Grained Execution Safety: Sandboxed shell execution with configurable read/write boundaries (standard and strict), interactive unified diff previews, and run budget caps.
  • Smart Shell Auto-Approval: A hybrid heuristic safety evaluator that auto-approves safe read-only and workspace commands, eliminating prompt fatigue while gating risky operations.
  • Multi-Agent Swarms: Spawn specialized foreground or background subagents (explorer, worker, mentor, librarian) across tiered model profiles (smart, balanced, cheap, chore).
  • Provider-Neutral Context Compaction: Optional compaction via /compact and agent.contextCompaction.enabled, preserving cold architectural context and hot-tail tool integrity. Session rollover is a separate, enabled-by-default handoff recommendation at token milestones.
  • Time-Travel Rewind & Forking: Non-destructive conversation rewinding with discard inspection (/rewind, /undo, /retry), session resumption (/resume, --resume), and branching (--fork).
  • Native Remote Development: Execute commands and manage codebases on remote servers seamlessly over SSH with local SSH agent authentication.
  • Private Web Gateway: Launch term2 serve to expose an authenticated Unix socket or TLS gateway for programmatic control and external clients.

How term2 Works

Section titled “How term2 Works”

When you launch term2 in a project directory:

  1. Context Discovery: term2 inspects the workspace root, detects project instructions (e.g., AGENTS.md, CLAUDE.md), discovers available skills, and establishes the operating profile.
  2. Interactive Run Loop: You interact via a composer supporting multi-line editing, @ file completion, ! direct shell execution, and / slash commands.
  3. Safe Tool Execution: When the agent proposes tool actions (such as editing files, running shell commands, or executing sandboxed JavaScript), term2 renders unified diffs or approval prompts before mutation occurs.
  4. Resilience & Resumption: All events are saved to append-only logs in your platform’s state directory, allowing full resumption or forking anytime.

Project Instructions and Skills

Section titled “Project Instructions and Skills”

Instructions are discovered from AGENTS.md or CLAUDE.md in the project and its parent directories. Skills are searched in project or user scope directories: .term2/skills, .agents/skills, and .claude/skills (under the project root or home directory). A skill is a directory containing a SKILL.md file.