opencode

opencode.ai · CLI Tool · Terminal

opencode is a CLI tool: AI coding agent built for the terminal. Its repository anomalyco/opencode carries 212,333 stars and 454 contributors, and is licensed MIT. The latest stable release is v1.18.35, published 6 Oct 2026 with binaries for linux, macos, windows. npm served 3,318,599 downloads of opencode-ai in the week we read. Anomaly's trust centre lists SOC 2 Type 1 and SOC 2 Type 2 as compliant, behind 3 internal policies and 35 monitored controls.

Stars: 212,333; Verified 8 Oct 2026
Last commit: 8 Oct 2026; 0 days old when we read it; Verified 8 Oct 2026
License: MIT; Verified 8 Oct 2026
Latest stable release: v1.18.35; published 6 Oct 2026; Verified 8 Oct 2026
Contributors: 454; Verified 8 Oct 2026
npm downloads: 3,318,599 a week; opencode-ai; Verified 8 Oct 2026
Homebrew installs: 21,432 in 30 days; opencode; Verified 8 Oct 2026
Repository: anomalyco/opencode; TypeScript; Verified 8 Oct 2026
Release binaries: linux, macos, windows; Verified 8 Oct 2026

Visit opencode anomalyco/opencode on GitHub opencode alternatives

What opencode is

OpenCode reaches 75+ LLM providers and local models from a single config file, because it resolves every model through the AI SDK and the Models.dev catalogue instead of a hand-written provider list.

“OpenCode uses the AI SDK and Models.dev to support 75+ LLM providers and it supports running local models.”

opencode.ai/docs/providers

What opencode costs

OpenCode Go, the optional subscription, costs $5 for the first month and $10 a month after that, and one workspace can hold only one subscriber.

“OpenCode Go is a low cost subscription — $5 for your first month, then $10/month — that gives you reliable access to popular open coding models.”

opencode.ai/docs/go

A workspace can hold only one Go subscriber, so a second developer needs a second workspace.

“Only one member per workspace can subscribe to OpenCode Go.”

opencode.ai/docs/go

Use opencode when

Explore an unfamiliar codebase before you change it: the built-in plan agent refuses file edits and asks before each bash command.

“**plan** - Read-only agent for analysis and code exploration · - Denies file edits by default · - Asks permission before running bash commands · - Ideal for exploring unfamiliar codebases or planning changes”

github.com/anomalyco/opencode

Run one task against two vendors' models in the same session: `/connect` adds a provider key and the model picker switches without a restart.

“Add the API keys for the provider using the /connect command.”

opencode.ai/docs/providers

Run the agent inside CI on your own runner: mentioning /opencode or /oc on a GitHub issue or pull request executes the task in GitHub Actions.

“OpenCode integrates with your GitHub workflow. Mention /opencode or /oc in your comment, and OpenCode will execute tasks within your GitHub Actions runner.”

opencode.ai/docs/github

Not the right fit for

A team that must lock the agent down before its first run should choose Codex CLI, which starts inside an OS-enforced sandbox with the network switched off.

“By default, the agent runs with network access turned off.”

developers.openai.com/codex/sandbox

What opencode is bad at

A fresh install edits files and runs shell commands with no approval prompt, because every permission except `doom_loop` and `external_directory` starts at allow.

“Most permissions default to "allow".”

opencode.ai/docs/permissions

Memory grows through a long terminal session: the maintainers opened a central memory thread on 2026-04-02, asked users for heap snapshots taken once the process reaches 1-2 GB, and the thread is still open 139 days later.

“We have scattered reports of memory issues so we are going to tackle them all in a central place.”

github.com/anomalyco/opencode

`/share` copies the transcript to OpenCode's servers, and the resulting link needs no password.

“Shared conversations are publicly accessible to anyone with the link.”

opencode.ai/docs/share

Security and data handling

Anomaly's trust centre lists SOC 2 Type 1 and SOC 2 Type 2 as compliant, behind 3 internal policies and 35 monitored controls.

“3 Policies · Internal policies that govern how we operate and protect customer data.”

trust.opencode.ai

Four named subprocessors touch the data — Amazon Web Services, GitHub, Google Workspace and Slack — and the Data Processing Agreement is released only under NDA.

“4 Subprocessors · Third parties that process data on our behalf under strict contractual safeguards. · Amazon Web Services”

trust.opencode.ai

Technical details

Category: CLI Tool · Terminal
Repository: anomalyco/opencode
Default branch: dev
Primary language: TypeScript
Repository created: 30 Apr 2025
Newest commit on the default branch: 8 Oct 2026
Package formats: appimage, deb, dmg, exe, rpm, tar.gz, zip
Architectures: arm64, x64
Registries checked: npm:opencode-ai,opencode, homebrew:opencode,opencode-desktop, vscode_marketplace:search(opencode)
Website check: https://opencode.ai answered HTTP 200

Alternatives to opencode

Claude CodeCommand-line AI coding agent that understands your codebase and automates tasks via natural languageCompare
AiderAI pair programming in your terminal with clean commitsCompare
Codex CLIOpenAI's lightweight terminal chat interface for code operationsCompare
GooseLocal on-machine AI agent that allows any LLM and MCP servers as extensionsCompare
MyCoder.aiOpen source AI coding assistant with Git/GitHub integration and parallel executionCompare

All opencode alternatives, ranked →

How we check this page

Machine facts. Stars, licence, the newest commit on the default branch, the latest stable release and download counts are read straight from the GitHub API and from npm, PyPI, Homebrew and the VS Code Marketplace. Each one shows the day it was last confirmed, not the day this page was built. A failed fetch keeps the old value and its old date rather than overwriting it.
Judgments. Every claim about what a tool suits, what it is bad at and what it costs is quoted from a page we read, and that page is linked beside the claim. If there is no link, we did not write the claim.
What we do not publish. No score out of five and no star rating. Every other claim here traces to a source, and a number we could not justify would undermine the ones that can.
This tool. Every claim carries the URL it came from and a verbatim quote from that URL, read on 2026-08-19.