OpenCode
OpenCode is an open-source, model-agnostic AI coding agent that runs as a terminal interface for questioning a codebase, planning features, and making or undoing code changes through a large language model of the user's choice. It connects to 75+ providers with the user's own API keys and targets developers who want provider independence.
Use Cases
Query and explain an unfamiliar codebase using file references
Plan a feature in read-only mode before editing code
Scaffold a new project or service from scratch
Implement a refactor or code change across multiple files
Undo AI-made edits during a session
Connect databases and external tools through MCP servers
Pros
Swaps between Claude, GPT, Gemini, and local Ollama without reconfiguration
Bring-your-own-key means you pay provider rates with no token markup
Most-starred open-source coding agent at roughly 180,000 GitHub stars
MIT license allows full inspection, modification, and self-hosting
Terminal UI pairs with LSP code intelligence and MCP tool connections
Cons
You manage your own API keys and track your own model costs
Windows support leans on WSL for acceptable performance
Desktop app is still in beta and self-hosted share pages are unfinished
Independent benchmarks rank Claude Code and Codex higher on task accuracy
Share links route conversations through OpenCode's CDN unless disabled
Platforms
Desktop
Plugin
Self-Hosted
MCP
OpenCode won the open-source coding-agent race on mindshare more than benchmark scores, and it did it by refusing to pick your model for you. Bring your own key and point it at whatever you trust that week, jumping from Claude to Gemini or a local Ollama build mid-project instead of being married to one vendor. You carry the flip side of that freedom, which is owning the token bill and tracking your own spend.
Must Try