OpenCode is Anomaly’s (formerly SST) open-source AI harness for the command line: a coding agent that makes language models operational through MCP tools, context management and a fine-grained permission system - freely combinable with more than 75 model providers. As of 20 September 2026.
What is OpenCode?
OpenCode grew out of SST (Serverless Stack); the sst GitHub organization’s own profile now points to its move to anomalyco, and the active repository lives at github.com/anomalyco/opencode, licensed under MIT per its LICENSE file. Per GitHub releases, the first versions shipped in May 2025. OpenCode runs as a terminal application (TUI), as a beta desktop app for macOS, Windows and Linux, and through a VS Code integration that installs automatically the first time it is launched from the terminal. It is installed via, among other options, npm i -g opencode-ai@latest, brew install anomalyco/tap/opencode, or the official install script, and started by running opencode inside a project directory.
OpenCode as an AI harness
The agent loop evaluates the model’s tool requests, executes them, and writes results back into the context - interactively in the terminal, headless via opencode run, or persistently via opencode serve with an OpenAPI interface. Tools are extended through MCP servers configured in opencode.json as type: "local" (a start command) or type: "remote" (a URL plus headers). Project rules live in AGENTS.md, falling back to CLAUDE.md for Claude Code migrations. Two built-in primary agents - build (full access) and plan (asks by default before file edits and bash commands) - are switched with the Tab key. Three built-in subagents handle sub-tasks: general (multi-step research and tasks with tool access), explore (read-only codebase search) and scout (read-only research on external docs and dependencies); custom subagents are defined as markdown files with YAML frontmatter in .opencode/agents/. Permissions are set per action (bash, edit, webfetch and others) to allow, ask or deny.
Models and model access
Per its documentation, OpenCode connects to more than 75 providers, drawing on the AI SDK and Models.dev - from Claude to GPT to Gemini, plus local models via Ollama, LM Studio or llama.cpp. With OpenCode Zen, the vendor also runs its own model marketplace curated for coding agents. Custom OpenAI-compatible endpoints are explicitly supported: in opencode.json you add a provider with "npm": "@ai-sdk/openai-compatible" and "options": {"baseURL": "..."}. That means OpenCode can be run through the AI Gateway, provided the gateway is registered as an OpenAI-compatible endpoint.
Enterprise use: data protection and governance
Per the documentation, OpenCode itself does not store code or context data; requests go locally or directly to the chosen model provider - the one exception being the optional /share feature, which creates public links and can be turned off via "share": "disabled". Organizations can plug their SSO into central configuration, and can also disable other model providers so that all requests route through approved infrastructure. Regardless of the harness used, source code and prompts go to whichever model provider is chosen, and that flow belongs in the record of processing activities beforehand.
Orchestrating OpenCode with herdr
Per herdr’s documentation, herdr supports OpenCode through a lifecycle plugin when installed, falling back to a screen manifest otherwise; the integration reports both state and session. Details and the full list of supported harnesses are on the herdr page.
Where it fits: developer harness and enterprise harness
OpenCode is a harness for developers working in the terminal - for business users who work without a command line, CompanyGPT is the matching harness, with MCP tools, enterprise knowledge, and roles sourced from Entra ID or Keycloak. The AI Gateway provides the shared cost and governance layer across both worlds. Our overview of all AI harnesses covers further coding agents and orchestrators. We help organizations integrate OpenCode and other coding harnesses cleanly into existing identity, cost and compliance structures.