A fork of OpenCode by EvalOps

This is a public fork maintained by EvalOps for internal use. We use OpenCode extensively and maintain this fork to experiment with enhancements aligned with our LLM evaluation workflows. OpenCode is open source and permissively licensed (MIT).

This fork tracks the dev branch of evalops/opencode. For official releases and documentation, see the upstream repository.

# YOLO
curl -fsSL https://opencode.ai/install | bash

# Package managers
npm i -g opencode-ai@latest        # or bun/pnpm/yarn
brew install sst/tap/opencode      # macOS and Linux
paru -S opencode-bin               # Arch Linux

The install script respects the following priority order for the installation path:

1. $OPENCODE_INSTALL_DIR - Custom installation directory
2. $XDG_BIN_DIR - XDG Base Directory Specification compliant path
3. $HOME/bin - Standard user binary directory (if exists or can be created)
4. $HOME/.opencode/bin - Default fallback

# Examples
OPENCODE_INSTALL_DIR=/usr/local/bin curl -fsSL https://opencode.ai/install | bash
XDG_BIN_DIR=$HOME/.local/bin curl -fsSL https://opencode.ai/install | bash

For more info on how to configure OpenCode head over to our docs.

You can inspect local usage history and tool telemetry with the built-in stats command:

opencode stats                   # pretty summary
opencode stats --json            # machine-readable output
opencode stats --telemetry all   # include recent tool runs
opencode stats --limit 50        # show more history
opencode stats --clear           # reset stored telemetry data
opencode stats --details         # show telemetry metadata fields
opencode stats --details-format ndjson --fields status,final_url
opencode stats --status error --since 1d
opencode stats --compare baseline.json --warn-latency 2000

Advanced telemetry usage tips:

• Capture a baseline for comparison with opencode stats --json --telemetry all --limit 500 > baseline.json, then diff with --compare baseline.json.
• Export metadata for dashboards using --details-format csv or --details-format ndjson.
• Focus on specific signals by pairing --status, --since, --until, and --fields filters.
• Gate builds by combining --warn-latency or --warn-errors with CI scripts.

The telemetry section lists recent tool executions (duration, status, error message) gathered from persisted tool.telemetry events.

OpenCode is an opinionated tool so any fundamental feature needs to go through a design process with the core team.

However we still merge a ton of PRs - you can contribute:

• Bug fixes
• Improvements to LLM performance
• Support for new providers
• Fixes for env specific quirks
• Missing standard behavior
• Documentation

Take a look at the git history to see what kind of PRs we end up merging.

To run OpenCode locally you need.

• Bun
• Golang 1.24.x

And run.

$ bun install
$ bun dev

API Client: After making changes to the TypeScript API endpoints in packages/opencode/src/server/server.ts, you will need the OpenCode team to generate a new stainless sdk for the clients.

It's very similar to Claude Code in terms of capability. Here are the key differences:

• 100% open source
• Not coupled to any provider. Although Anthropic is recommended, OpenCode can be used with OpenAI, Google or even local models. As models evolve the gaps between them will close and pricing will drop so being provider-agnostic is important.
• A focus on TUI. OpenCode is built by neovim users and the creators of terminal.shop; we are going to push the limits of what's possible in the terminal.
• A client/server architecture. This for example can allow OpenCode to run on your computer, while you can drive it remotely from a mobile app. Meaning that the TUI frontend is just one of the possible clients.

The other confusingly named repo has no relation to this one. You can read the story behind it here.

Join our community Discord | X.com