OpenCode

Name: OpenCode
Author: AgentRadar Editorial Team

Verified

Multi-provider CLI with build/plan agent modes and 75+ LLM support

Coding Agents Open Source

8.8 /10

Visit Website GitHub

Released

2025-04

Country

United States

API

Available

Self-Host

Yes

GitHub Stars

176,000

Funding

Series B

Last Reviewed

2026-06

About OpenCode

OpenCode is the most provider-flexible coding agent available — an open-source, MIT-licensed tool from SST that connects to 75+ LLM providers through Models.dev, including frontier models, regional providers, and fully local models. It ships as a CLI, desktop app (macOS, Windows, Linux), and IDE extension via the Agent Client Protocol, so it meets developers wherever they work. Its agent architecture is its differentiator: a "build" agent has full filesystem and terminal access for hands-off implementation, while a "plan" agent operates read-only for analysis and exploration without risk. A "general" subagent handles complex multi-step tasks that exceed a single agent's scope. Multi-session support lets parallel agents work on different tasks simultaneously. LSP auto-loading provides language-aware diagnostics, share links let teams exchange conversations, and integrations with GitHub Copilot and ChatGPT bridge the gap between OpenCode and existing tooling. With 176K+ GitHub stars, 900+ contributors, and 7.5M claimed monthly developers, it is one of the fastest-growing open-source projects in the coding agent space.

Our Verdict

The most provider-flexible coding agent in the ecosystem — 75+ LLMs, MIT-licensed, and available as CLI, desktop, and IDE extension. The build/plan agent architecture is a thoughtful design that lets developers control access levels per task. If you want maximum model choice, local model support, and a growing open-source community, OpenCode is the best pick.

Features

75+ LLM provider support via Models.dev

Build agent (full-access) and plan agent (read-only) modes

General subagent for complex multi-step tasks

Multi-session parallel agents

LSP auto-loading for language-aware diagnostics

Desktop app (macOS, Windows, Linux)

Agent Client Protocol (ACP) for IDE integration

Share links for team collaboration

Detailed Ratings

Ease of Use

8.0

Value for Money

9.5

Features

9.0

Support

8.0

Performance

9.0

Overall Rating

8.8 /10

Pros & Cons

Pros

Unmatched provider flexibility — 75+ LLMs including local and regional models
Build/plan agent modes let you control access level per task — safe exploration or hands-off implementation
Open-source MIT license — no vendor lock-in, fully auditable
Runs everywhere: CLI, desktop, and IDE via Agent Client Protocol
176K+ stars and massive contributor base — one of the fastest-growing dev tools

Cons

Managing accounts and API keys across 75+ providers can be complex
Rapid growth means documentation and stability can lag behind features
Younger project than Claude Code or Copilot — patterns and best practices still evolving

Use Cases

Multi-provider CLI development with model choice per taskRead-only codebase analysis and exploration with plan agent modeParallel agent sessions for tackling multiple tasks simultaneouslyCost optimization by routing to the best-value model for each workloadTeams wanting share links and collaborative agent sessions

Who Is It For?

Developers who want maximum model flexibility across 75+ providers with build/plan agent modes and a choice of CLI, desktop, or IDE

#cli#desktop#open-source#multi-provider#agents#acp

Pricing Plans

Open Source

Free

Full CLI, desktop, and IDE extension
75+ LLM providers supported
Build and plan agent modes
MIT license

LLM API Costs

Usage-based /BYOK

Pay your chosen LLM provider (75+ supported)
Free models available via Models.dev
Local models via Ollama are free

Frequently Asked Questions

Is OpenCode really free?

Yes — OpenCode is MIT-licensed and fully free. You pay only for the LLM API keys you bring, and you can choose from 75+ providers including free or local models to minimize cost.

What are build and plan modes?

Build mode gives the agent full filesystem and terminal access to implement changes autonomously. Plan mode operates read-only — the agent can explore, analyze, and reason about your codebase but cannot modify anything. This makes plan mode safe for exploration and build mode for execution.

How is OpenCode different from Claude Code?

OpenCode is model-agnostic with 75+ provider support and offers both CLI and desktop interfaces. Claude Code is Anthropic-first with a 1M-token context window. OpenCode's strength is provider flexibility and the build/plan agent architecture; Claude Code's strength is frontier Claude model access and deep codebase context.

What is the Agent Client Protocol?

ACP is an open protocol that lets OpenCode integrate into IDEs as an agent. Currently it powers the Zed integration, and other IDEs can adopt it. It is conceptually similar to MCP but for agent-to-IDE communication rather than agent-to-tool.