Kenhuang Academy
  • English

      • Introduction to OpenCode

      What is OpenCode?

      OpenCode is an open-source AI coding agent maintained by Anomaly, available in the terminal (TUI), desktop app, and IDE extensions. Unlike single-vendor assistants, OpenCode supports 75+ LLM providers via Models.dev—including Claude, GPT, Gemini, local models, and OpenCode’s own Zen model gateway.

      Core value:

      • Model-agnostic: one agent workflow; switch providers by changing config
      • Privacy-first: official stance is no storage of your code or context (local/self-hosted providers keep data paths under your control)
      • LSP integration: connect language servers and feed diagnostics back to the agent
      • Parallel sessions: run multiple agent sessions in the same project
      • Open source: actively maintained at anomalyco/opencode

      Product Matrix

      ┌─────────────────────────────────────────────────────────┐
│              OpenCode (open-source agent core)           │
└─────────────────────────────────────────────────────────┘
     ┌──────────────┬──────────────┬──────────────┐
     │  Terminal    │   Desktop    │  IDE ext.    │
     │  TUI / CLI   │  macOS/Win   │  VS Code etc.│
     └──────────────┴──────────────┴──────────────┘
     opencode          Desktop beta      Plugin form
     opencode run      Same              Same
     opencode serve    —                 —
      EntryCommand / methodUse case
      TUIopencodeDaily interactive dev, Plan/Build switching
      One-shotopencode run "..."Scripts, CI, automation
      Headlessopencode serveRemote API, shared team instance
      Web UIopencode webBrowser access to local server
      DesktopDownload from siteDevelopers who prefer GUI

      Built-in Agents Overview

      OpenCode splits work between Primary Agents and Subagents:

      AgentTypeRole
      BuildPrimaryDefault dev mode, full tools
      PlanPrimaryRead-only planning; edits/commands need approval
      GeneralSubagentComplex multi-step tasks, can run in parallel
      ExploreSubagentFast read-only codebase exploration
      ScoutSubagentExternal docs and dependency source lookup

      Press Tab to switch Primary agents; invoke Subagents manually with @explore and similar.

      Comparison with Codex and Claude Code

      DimensionOpenCodeCodexClaude Code
      Open source✅CLI open sourceCLI open source
      Model choice75+ providersOpenAI-focusedAnthropic-focused
      Project instructionsAGENTS.md (via /init)AGENTS.mdCLAUDE.md etc.
      Planning modePlan agent (write disabled)Read-only / approval modesPlan mode
      LSP✅ Built-in multi-languageDepends on IDE/toolchainDepends on integration
      Session sharing/share linkCloud/PRProduct-dependent
      PricingFree OSS + pay APIChatGPT plan / APIAPI / subscription

      When to choose:

      • Switch models often or run local LLMs → OpenCode
      • Deep OpenAI / ChatGPT stack → Codex
      • Deep Anthropic stack → Claude Code

      All three support the AGENTS.md open standard; project conventions can be partially reused.

      Good Fit / Caveats

      Good fit:

      • Multi-model comparison and cost tuning (cheap model for Plan, strong model for Build)
      • Privacy-sensitive setups (self-hosted providers, local models)
      • LSP diagnostics to help the agent fix type/lint errors
      • Parallel sessions (e.g. one for frontend, one for API)

      Watch out:

      • Without API keys and without OpenCode Zen, you must wire up models yourself
      • Native Windows support is still maturing; WSL is smoother
      • LSP is optional; enabling everything on huge repos may use a lot of memory

      Billing and Providers

      • OpenCode software is free and open source
      • Model usage is billed by your chosen provider (OpenAI, Anthropic, etc.)
      • OpenCode Zen: curated model list, connect via /connect
      • Some providers may reuse GitHub Copilot, ChatGPT Plus, etc. (check current official support)

      Next Steps