OpenCode

OpenCode is an open-source AI coding agent from Anomaly, the team behind SST. It helps developers read, edit, refactor, test, and navigate codebases through a terminal interface, desktop app, or IDE extension. The project is MIT-licensed and developed in the open at the anomalyco/opencode GitHub repository.

OpenCode sits in the same broad category as Codex, Claude Code, Cursor agents, and GitHub Copilot agent mode, but with a strong terminal-first and provider-flexible design. You can connect it to hosted providers such as OpenAI, Anthropic, Google, DeepSeek, OpenRouter, GitHub Copilot, GitLab Duo, and many others, or run it against local models through compatible providers.

What OpenCode Does

Capability	How OpenCode handles it
Codebase questions	Reads files, searches code, and uses project context to explain how a codebase works.
Code changes	Uses edit, write, and apply_patch tools to modify files inside the project.
Shell tasks	Can run terminal commands through its bash tool when permissions allow it.
Planning	Supports a plan-style workflow before switching into build mode for changes.
Multi-session work	Can run multiple sessions or agents against the same project.
Provider choice	Supports 75+ LLM providers through Models.dev and the AI SDK, including local models.
MCP tools	Can connect local or remote MCP servers so external tools become available to the agent.
Team sharing	Supports share links for sessions when developers want to debug or hand off context.

Permissions and Safety

OpenCode gives developers control over what the agent can do through its permission configuration. Actions can be allowed, denied, or set to ask for approval. That matters because coding agents operate inside real repositories: reading files is low risk, but editing code, applying patches, running shell commands, or calling external tools can have real consequences.

The permission system lets teams decide whether tools such as bash, edit, write, webfetch, and MCP tools should run automatically or require human approval. For example, a team might allow code search, require approval for shell commands, and deny destructive commands by pattern. This makes OpenCode more suitable for professional development workflows than a raw chatbot connected to a terminal.

Why Developers Use It

OpenCode is useful when developers want an AI coding agent that is open-source, auditable, local-workflow friendly, and not tied to one model provider. That makes it attractive for teams that want to experiment with different models, use existing subscriptions, connect internal tools, or avoid putting all coding-agent usage behind a single proprietary interface.

Its main tradeoff is the same one shared by most open coding agents: flexibility creates configuration work. A hosted proprietary agent may be smoother out of the box. OpenCode gives more control over providers, tools, permissions, prompts, MCP servers, and local workflow integration, but that control is only valuable if the team is willing to configure and maintain it.

OpenCode vs Codex and Copilot

OpenCode is not a model. It is an agent harness around models. Codex and GitHub Copilot are vertically integrated products from OpenAI and GitHub. OpenCode is closer to an open, model-agnostic workbench: the value is the interface, tool layer, permissions, provider support, and ability to run the same coding workflow across many models.

The short version: OpenCode is for developers who want a coding agent they can inspect, configure, and plug into their own model and tool stack.