OpenCode

OpenCode is a terminal-based AI coding assistant that operates as a TUI (Terminal User Interface) application, distinguishing itself from IDE-plugin competitors by living entirely in the command line. It connects to any LLM provider through configurable API keys, giving developers the freedom to choose their preferred model without vendor lock-in. Supported providers range from major platforms to specialized services including 302.AI, Baseten, Cerebras, Snowflake Cortex, and Helicone, with configuration managed through environment variables and JSON settings files.

A defining feature is Plan mode. Before OpenCode writes any code, Plan mode lets it propose an implementation strategy without modifying files. The documentation recommends asking the agent to create a plan before committing to changes, turning what could be an opaque code-generation sprint into a reviewable design conversation. This editorial-style workflow — propose, review, then execute — aligns with how senior developers prefer to work with AI assistance.

OpenCode Zen, the project's optional first-party AI provider, offers a pay-as-you-go model charged per million tokens. Users sign in, add billing details, and connect via the /connect command in the TUI. Zen is explicitly optional — developers can bring their own API keys from other providers instead. This dual approach means OpenCode can serve both teams that want a managed AI backend and those that already have provider relationships in place.

The Enterprise tier adds SSO integration, central configuration management, and a notable data privacy commitment: OpenCode does not store code or context data. All processing happens locally or through direct API calls to the chosen AI provider. The one exception is the optional /share feature, which sends conversation content to the project's hosting infrastructure when users explicitly create public share links. Enterprise pricing follows a contact-sales model after internal team trials.

Conversation sharing supports three modes — manual, auto-share for every new conversation, and disabled — with clear privacy guidance recommending against sharing conversations containing proprietary code or confidential data. Shared conversations generate unique URLs for team collaboration or external help requests.

The theme system uses a flexible JSON format with truecolor (24-bit) support, allowing developers to customize syntax highlighting, UI elements, and gray scales based on terminal background color. Without truecolor, themes fall back to 256-color approximation. Keybindings are likewise configurable through JSON, with platform-specific defaults for Windows terminals that do not support POSIX suspend.

Installation is available via package managers, though Windows support using Bun is documented as in progress, signaling that the primary development target remains Unix-like environments for now.

For teams already working in the terminal — developers who live in tmux, vim, and CLI workflows — OpenCode fits naturally into the existing toolchain rather than demanding a switch to a specific IDE. The multi-provider architecture also means it can adapt as the LLM landscape shifts. Explore more tools in the AI Agents Directory.