Tools· Software Development & Technical Infrastructure

    OpenCode

    Open source AI coding agent for the terminal, IDE, and desktop that analyzes projects through an init routine and works with either your own LLM provider keys or a curated model marketplace.

    generatingautomatingexecutableOpen Source

    Description

    Strengths

    Open source core with multiple access paths
    Terminal, IDE integration, and a desktop app in beta are all available for the same agent.
    Broad choice of installation methods
    An install script plus native package managers for macOS, Linux, and Windows, including Homebrew, Scoop, Chocolatey, Pacman, AUR, mise, and Nix.
    Separate agents for planning and implementation
    A full-access mode for implementation and a read-only mode for analysis and planning, switchable with the Tab key, plus a general subagent for multistep tasks.
    Automatic project onboarding
    An init routine analyzes the codebase and creates a versionable context file with structure and code patterns.
    Traceable undo
    An unwanted change can be discarded, the original prompt is preserved and can be adjusted directly.

    Assessment

    AI features

    • Plan mode as a safety net Denies file edits by default and asks for permission before every bash command, before the full-access mode takes over implementation.
    • General subagent for multistep tasks For complex, multistep searches and tasks, invokable directly through a mention in a message.
    • Curated model access through Zen Hand-picked models that have been tested and benchmarked for coding agents, as an alternative to using your own provider keys, currently available through a waitlist for early access according to the provider's website.
    • Image understanding through drag and drop Screenshots and images can be dragged directly into the terminal, OpenCode recognizes them and factors them into its response.

    Suitable for

    • Development teams that want an open source coding agent in the terminal or IDE and want to choose their own LLM provider
    • Organizations with compliance requirements that want to bind code and context data to their own infrastructure through central SSO and AI gateway configuration (OpenCode Enterprise)
    • Individual developers and small teams that want to get started with their own API keys without building their own infrastructure, alternatively with the curated Zen access, for which the provider currently runs a waitlist for early access
    • Less suitable for Less suitable for Windows teams that depend on a fully finished Bun-based installation, which was still described as in progress by the provider

    Limitations and notes

    • Windows installation through Bun not yet finished For the best Windows experience, the provider currently recommends a different installation path; the Bun-based installation is still in progress according to the documentation.
    • Provider access is a prerequisite Without a configured LLM provider API key or Zen access, and a modern terminal emulator, OpenCode cannot be used productively.
    • Shared conversations run through a CDN Shared links are served through the provider's own CDN and cached there; for enterprise trials, the provider recommends disabling this feature for now.
    • Self-hosting of share pages is still on the roadmap Running the sharing pages fully on your own infrastructure is, according to the provider, only planned for now, not a finished feature.
    • Private npm registries require your own login For projects with private package registries, you must be logged into the relevant registry through Bun before starting OpenCode.

    Quick start

    1. Install OpenCode through the install script or a supported package manager.
    2. Sign in with an LLM provider, add billing details, and copy the API key, or alternatively select a different provider and enter its key.
    3. Start OpenCode in the desired project directory and run the init routine so the agent picks up the project's structure and code patterns.
    4. In read-only mode, describe the planned change and give feedback, then switch to full-access mode with the Tab key and kick off the implementation.
    5. If needed, share the conversation as a link, or discard an unwanted change and adjust the prompt.

    Tips

    • For unfamiliar codebases, start with read-only mode; it denies file edits and asks for permission before every bash command.
    • Version the context file created by the init routine so the whole team benefits from the captured project structure.
    • Drag images and screenshots directly into the terminal instead of describing them.
    • For enterprise trials, disable the CDN-backed sharing feature for conversations at first and instead use the central SSO configuration for your own AI gateway.

    Access

    Last reviewed: · Pricing, plans and features are a snapshot in time. Check the provider's own page before deciding.

    In the workshop this becomes your method.

    Whoever sees this process run once wants the agent behind it next. We build that in the workshop From Process to Agent.

    View workshops

    Related resources

    Browse all resources

    Conversation, not pitch

    Understand first, then decide. We take time for an initial conversation, without sales pressure, without obligation.

    Schedule a call