opencode
The terminal-native agent with the largest open-source community in the category.
opencode (MIT) is a coding agent that runs in your terminal, with a desktop app in beta and IDE extensions alongside it. It is deliberately provider-agnostic — 75+ providers via the AI SDK and Models.dev, including Ollama, LM Studio and any OpenAI-compatible local endpoint — and it ships two built-in agents: a full-access `build` agent and a read-only `plan` agent for exploring a codebase before anything is allowed to change it. It is now the most-starred project in this space at roughly 191k GitHub stars, maintained by Anomaly after the repository moved from `sst/opencode` to `anomalyco/opencode`. The optional opencode Zen gateway sells curated, benchmarked models pay-as-you-go and includes some free ones; using your own provider keys stays free.
What it does well
- +MIT and provider-agnostic by design — 75+ providers, plus Ollama, LM Studio or any OpenAI-compatible local endpoint
- +The read-only `plan` agent lets you scope work before the agent can touch a file
- +Terminal-first, so it works over SSH on a server or a modest machine
- +The largest community in the category and daily commits
Where it falls short
- −Terminal-first is a genuine adjustment coming from an editor-integrated agent; the desktop app is still beta
- −No inline autocomplete — a different tool for a different job than Cursor's Tab
- −Ownership moved from SST to Anomaly; the MIT licence means no practical risk, but it's worth knowing
- −Large configuration surface (providers, models, agents) — more to learn than picking from a dropdown
Hardware note
The client is a single binary and runs comfortably on modest hardware, including over SSH. Running the model locally is the demanding part: budget ~24 GB VRAM or a 32 GB+ unified-memory machine for a 30B-class quantised coding model, and expect a capability gap against a frontier cloud model on multi-step agent work.
opencode as an alternative to
Where opencode shows up in our comparisons, and how it ranked.
opencode head-to-head
Straight comparisons against the tools people weigh it against.