OpenCode
Overview
Section titled “Overview”| Company | SST |
| Type | Terminal CLI (TUI) |
| Pricing | Free — bring your own API key |
| Protocol | MCP |
| Config Formats | AGENTS.md |
| License | MIT |
| Website | opencode.ai |
What It Does
Section titled “What It Does”OpenCode is a terminal coding agent built around one idea: the harness should not be tied to a model vendor. You bring the provider — Anthropic, OpenAI, Google, or a local model — and the agent, the TUI, and the workflow stay the same.
It runs as a full terminal UI rather than a line-by-line REPL, with a client/server split that lets the agent run somewhere other than the terminal you are watching it from.
Key Strengths
Section titled “Key Strengths”- Provider-agnostic: Switch models without switching tools or relearning the workflow
- MIT licensed: Fork it, audit it, run it in environments that forbid closed agents
- Terminal-native TUI: Built for people who live in the terminal, not a stripped-down chat box
- Client/server split: Drive a remotely running agent from a local terminal
- Local models: Works with self-hosted and open-weight models for privacy-bound work
Trade-offs
Section titled “Trade-offs”- Quality tracks whichever model you point it at, so a weak provider produces a weak agent
- No first-party cloud agent or managed sandbox
- Bring-your-own-key means you handle cost control yourself
Where to find skills for OpenCode
Section titled “Where to find skills for OpenCode”A skill is a SKILL.md file, so it does not belong to a model vendor any more than OpenCode does. The same skill folder keeps working when you switch providers.
The largest open catalogue is atskills.one, with 60,000+ agent skills. See @skills for how referencing and saving work.
Best For
Section titled “Best For”- Teams that want one agent workflow across several model providers
- Privacy-sensitive work on local or self-hosted models
- Developers who want an open-source harness they can audit and modify