OpenCode: The Universal Open-Source AI Coding Agent

OpenCode is an open-source AI coding agent that works across terminals, IDEs, and desktop apps. It provides access to over 75 LLM providers and integrates with existing subscriptions like GitHub Copilot while maintaining a strict privacy-first policy. With features like LSP support and multi-session capabilities, it serves over 5 million developers as a versatile alternative to proprietary AI tools.

Key Points

• OpenCode is a cross-platform, open-source AI agent available in terminals, IDEs, and as a desktop app.
• It supports over 75 LLM providers and allows users to bring their own subscriptions like GitHub Copilot or ChatGPT Plus.
• The tool is privacy-focused and does not store any user code or context data, making it safe for sensitive environments.
• Advanced features include LSP integration, multi-session parallel agents, and shareable links for debugging sessions.
• The project has significant community backing with 120k GitHub stars and 5 million monthly active developers.

Sentiment

The community is predominantly critical of OpenCode despite enthusiasm for the open-source concept. Security and privacy concerns dominate, with commenters uncovering that the tool's actual behavior contradicts its privacy-first marketing. Development quality issues compound the skepticism, though some users still value having an open-source alternative to proprietary coding agents.

In Agreement

• OpenCode fills an important niche as an open-source, provider-agnostic alternative to proprietary coding agents like Claude Code and Codex
• The flexibility to connect to 75+ LLM providers and use existing subscriptions is genuinely valuable for developers
• Multi-session and LSP support are meaningful technical differentiators that improve the coding agent experience
• Having an open-source option matters for developers working in environments that require self-hosted models

Opposed

• OpenCode silently sends prompts to external services (Grok free tier) even when users configure local models, violating the privacy-first claims
• The project's authentication mechanism allows provider websites to supply arbitrary shell commands for execution, creating a serious RCE vulnerability
• Development practices are reckless: extremely high release cadence with constant breakage, no proper testing, and no changelogs
• The TypeScript codebase is bloated and resource-inefficient, often consuming over 1GB of RAM for what is essentially a terminal application
• The framing as privacy-first is misleading when the default configuration routes data through services that train on user inputs
• Agentic coding tools like OpenCode produce codebases that lack coherent architecture and are not fully understood by their developers