OpenCode is an open-source agentic coding tool from the team behind SST (Serverless Stack). It runs in the terminal as a keyboard-driven TUI (text user interface) and connects to multiple LLM providers, giving the model read/write access to the local file system and shell. Released under the MIT license, it is designed as a fully open alternative to proprietary coding CLIs.
Key capabilities
Terminal UI (TUI) — OpenCode provides a rich terminal interface with panes for the conversation, file diffs, and the working directory tree. Navigation is keyboard-driven, making it fast to use without a mouse.
Multi-provider LLM support — Connects to Anthropic Claude, OpenAI GPT-4o, Google Gemini, DeepSeek, Llama (via Ollama), and other providers. The model is selected per session, allowing cost/quality trade-offs.
File system and shell access — The agent reads, creates, edits, and deletes files within the project and runs shell commands (build tools, test runners, git operations).
Git integration — Applies changes to files and optionally commits them to the current branch with auto-generated messages.
Session persistence — Conversation history and session state are persisted to disk, allowing developers to resume a task across terminal sessions.
Autonomy level
Level 3 (supervised agent): the developer reviews proposed file edits and confirms tool calls before execution. All actions are visible in the TUI before they run.
Strengths
- MIT license — fully open source, no usage restrictions
- Model-agnostic with first-class support for multiple providers
- Rich TUI reduces context switching compared to plain text interfaces
- Session persistence supports long-running, multi-session tasks
Limitations
- Smaller community and ecosystem than Aider or Claude Code
- TUI requires a compatible terminal emulator; some CI environments lack one
- Less mature than longer-established competitors