AI-IDE, LLM-based IDE Opencode

AI-IDE, LLM-based IDE Opencode

2026-02-23. Category & Tags: AIGC, GPT, LLM, Large-Language-Model, 大语言模型, AI-IDE

See also:

OPENCODE (CLI) #

免费教程:

ps: OpenCode can also be integrated into GitHub and GitLab, see: opencode.ai: e.g. run opencode github install in a local repo.

WARN !!!注意!!! OpenCode 在 GPT 模型上会出现不使用缓存的问题,导致超高额度消耗。

install opencode cli #

npm i -g opencode-ai

frequently used slash-commands #

/timeline # 查看当前对话(session)的聊天记录,可以回退(revert)到选中的聊天记录执行前的状态

suggested config #

AI-toolbox is suggested.

Add the content below (disable Ctrl+z bug): image.png

{
  "keybinds": {
    "terminal_suspend": "none"
  }
}

ref reason

create our own slash-commands/agents #

commands for either build/plan mode; agents as either primary/subagent.

ref: video Bilibili by 技术爬爬虾 OpenCode详细攻略,开源版Claude Code,免费模型与神级插件

插件推荐 之 oh my opencode (omo) #

DEF OMO = 预置常用工具 + 预置常用 MCP + 预置 Agent

also: 老的 superpowers 2026-01 https://www.bilibili.com/video/BV1qm6bB5EL3/ 对比较多,步骤不清楚,不算教程

install: #

option 1 (suggested):

Then use AI-Toolbox to manage LLM providers and plugins.

option 2 (NOT suggested):

open opencode, then paste:

Install and configure oh-my-opencode by following the instructions here:
https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md 

and enter.

ref: github

ps: install/activate/use opencode (besides claude/codex etc.) in github. e.g. ref: video Bilibili by 技术爬爬虾 在Github利用Actions自动调用OpenCode修复 issues

typical config: #

config file: ~\.config\opencode\oh-my-opencode.json , or .jsonc.

suggested: use AI-toolbox to manage omo config:

点“新建配置” image.png

输入配置名称,例如“config1”,点“json 导入”: image.png 会自动解析以下 4 个方面的配置: 1.主 Agents(协调和决策): image.png 2.子 Agents(领域专家,被主 Agents 调用): image.png|697

3.按任务类别分类配置: image.png 4.其他配置: image.png

we can also modify the config file (~/.config/opencode/oh-my-opencode.json) directly (NOT suggested), example:

with mixed providers:

{
  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
  "google_auth": false,
  "agents": {
    "Sisyphus": {
      "model": "claude/opus-4.6"
    },
    "oracle": {
      "model": "openai/gpt-5.4"
    },
    "librarian": {
      "model": "opencode/glm-5"
    },
    "explore": {
      "model": "google/antigravity-gemini-3-flash"
    },
    "frontend-ui-ux-engineer": {
      "model": "google/antigravity-gemini-3-pro-high"
    },
    "document-writer": {
      "model": "google/antigravity-gemini-3-flash"
    },
    "multimodal-looker": {
      "model": "google/antigravity-gemini-3-flash"
	}
  }
}

with pure GPT:

{
  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
  "google_auth": false,
  "agents": {
    "atlas": {
      "model": "the_provider/gpt-5.4-xhigh"
    },
    "explore": {
      "model": "the_provider/gpt-5.4-xhigh"
    },
    "hephaestus": {
      "model": "the_provider/gpt-5.3-codex"
    },
    "librarian": {
      "model": "the_provider/gpt-5.4-xhigh"
    },
    "metis": {
      "model": "the_provider/gpt-5.4-xhigh"
    },
    "momus": {
      "model": "the_provider/gpt-5.4-xhigh"
    },
    "multimodal-looker": {
      "model": "the_provider/gpt-5.4-xhigh"
    },
    "oracle": {
      "model": "the_provider/gpt-5.3-codex"
    },
    "prometheus": {
      "model": "the_provider/gpt-5.4-xhigh"
    },
    "sisyphus": {
      "model": "the_provider/gpt-5.4-xhigh"
    }
  },
  "categories": {
    "unspecified-high": {
      "model": "the_provider/gpt-5.3-codex"
    },
    "unspecified-low": {
      "model": "the_provider/gpt-5.3-codex"
    },
    "visual-engineering": {
      "model": "the_provider/gpt-5.4-xhigh"
    },
    "writing": {
      "model": "the_provider/gpt-5.4-xhigh"
    }
  }
}

ref

ps: 必须配置 Sisyphus Prometheus Hephaestus Atlas (as of 2026-03);其他辅助适当配置。

现有问题: #

  1. omo 需要重视示例,现在对于为什么给示例不了解,让它照着极简示例写,结果写出 maven 等大的架构,而且一堆没用的中间文件(例如下载依赖用的脚本、非当前系统需要的文件等等)。
  2. 接上条,大问题需要问,不清楚的重要问题(例如需要工程级别还是 demo 级别)竟然不问,是否需要当做极其复杂的演进工业项目,还是只要仿造当前 demo,没有先 (根据给的例子) 判断,也不问。
  3. 每次做完一项任务竟然不自动 git commit,即使是单一 master branch,因为 system_prompt 优先级高于 agents.md。
  4. 如果有在 git repo, move, rename 需要自动用 git mv
  5. 网络下载持续低速需要特殊配置
  6. rate limit 等很多场景不能 fallback,需要使用“伴侣插件”@azumag/opencode-rate-limit-fallback

omo 插件的“伴侣插件” #

@azumag/opencode-rate-limit-fallback image.png

此配件不检查用户目录,配置文件 rate-limit-fallback.json 需要放到项目根文件夹或项目下 ./.config/ 等子文件夹。

WARN:插件注意写全称,从 @ 开始,且配置文件前缀与插件名称不同!

配置示例:

{
  "enabled": true,
  "cooldownMs": 60000,
  "fallbackMode": "cycle",
  "fallbackModels": [
    { "providerID": "siliconflow-cn", "modelID": "glm-5" },
    { "providerID": "openrouter-glob",    "modelID": "gemini-2.5-pro" },
    { "providerID": "openai_gpt",    "modelID": "gpt-5.4" }
  ]
}

问题

  • 不能支持 A\ API 格式(cxcc 不可)
  • 并且仍然有不会 fallback 的情况!
    • 扣费失败
    • 服务部署已超过最大限制(超单位时间限额)

ps: 可以用软链接,有问题只需修改用户目录的配置文件,在项目根目录运行(cmd ps 通用): cmd /c mklink "rate-limit-fallback.json" "%USERPROFILE%\.config\opencode\rate-limit-fallback.json"

omo features quick overview: #

  • Agents:
    • (Metis: 前置分析师,主 Agent 的顾问,先确定到底要做什么,有什么坑)
    • Sisyphus: the main agent, 先决定自己干还是专家组队干,组队时负责任务分类、任务委派、全局协调 (claude-opus-4-6 / kimi-k2.5 / glm-5 )
    • (Momus: 计划质检员,给 Sisyphus 挑刺,确保计划可执行、无遗漏)
    • Prometheus: planner, 拆分工作,生成依赖图和计划,不管具体代码 (gpt-5.4 / claude-opus-4-6 / kimi-k2.5 / glm-5 )
    • Hephaestus: deep agent, autonomous worker, 自主决策的工匠,先 10 分钟计划,然后埋头苦干 (gpt-5.3-codex)
    • Atlas: Plan Executor, 任务管理者,持有 todo-list 的“项目经理“,跟踪多步骤任务,确保有序完成不遗漏(被自动调用)
    • Oracle: architecture/debugging, CTO 的战略顾问,分析审查调试,只看不动手 (gpt-5.3-codex)
    • Librarian: docs/code search, 读文档、搜代码、找开源 (glm-5 )
    • Explore: fast codebase grep, 找函数:哪里定义、哪里用了 (gemini-3-flash)
    • Multimodal Looker, multimedia, 多媒体视觉分析,图片、PDF、PPT 等 (gemini-3-flash)
    • frontend-ui-ux-engineer (gemini-3-pro)
    • document-writer (gemini-3-flash)
  • Background Agents: Run multiple agents in parallel like a real dev team
  • LSP & AST Tools: Refactoring, rename, diagnostics, AST-aware code search
  • Context Injection: Auto-inject AGENTS.md, README.md, conditional rules
  • Claude Code Compatibility: Full hook system, commands, skills, agents, MCPs
  • Built-in MCPs: websearch (Exa), context7 (docs), grep_app (GitHub search)
  • Session Tools: List, read, search, and analyze session history
  • Productivity Features: Ralph Loop, Todo Enforcer, Comment Checker, Think Mode, and more

(pic ref: AI 超元域)

Main agent: Sisyphus (Opus 4.5 High), which uses the tools below:

  • Sisyphus’s Teammates (Curated Agents)
    • Hephaestus: Autonomous deep worker, goal-oriented execution (GPT 5.2 Codex Medium) — The Legitimate Craftsman
    • Oracle 先知: Design/architecture, debugging (GPT 5.2 Medium)
    • Frontend UI/UX Engineer: Frontend development (Gemini 3 Pro)
    • Librarian 图书馆管理员: Official docs, open source implementations, codebase exploration (Claude Sonnet 4.5)
    • Explore 网络搜索: Blazing fast codebase exploration (Contextual Grep) (Claude Haiku 4.5)
  • Full LSP / AstGrep Support: Refactor decisively.
  • Todo Continuation Enforcer: Forces the agent to continue if it quits halfway. This is what keeps Sisyphus rolling that boulder.
  • Comment Checker: Prevents AI from adding excessive comments. Code generated by Sisyphus should be indistinguishable from human-written code.
  • Claude Code Compatibility: Command, Agent, Skill, MCP, Hook(PreToolUse, PostToolUse, UserPromptSubmit, Stop)
  • Curated MCPs:
  • Exa (Web Search)
  • Context7 (Official Documentation)
  • Grep.app (GitHub Code Search)
  • Interactive Terminal Supported - Tmux Integration
  • Async Agents

ref: video Bilibili by 技术爬爬虾 OpenCode详细攻略,开源版Claude Code,免费模型与神级插件

omo experience 使用技巧/经验分享: #

ulw ultra-work, for multi-agent to work for different models in parallel.

2026-01 OpenCode + Oh My OpenCode

建议配置好:自动 /handoff ,避免只根据文字压缩上下文。

评论区 of 一人抵一个开发团队!OpenCode最强插件Oh My OpenCode, AI超元域 2026-01 (文本教程, OMO中级(进阶)水平)

by 硬币大仙人: 这俩我都用,简单点说 claude code 要自己折腾配置才好用, omo 插件加持下的 opencode 基本上是开箱即用,就是废订阅。我也尝试过改 Sisyphus 的 model minimaxm2.1 自我感觉不行,主力还是得靠 opus4.5。 上来先用计划 agent 制定 “agents.md”,调教好以后,小任务直接两三句就交给他做,复杂点的任务再套一层 ralph-loop 配合他自己内置的 continue hook,两层续写,基本可以做到交代完之后 就 hands-off。 omo 有一个很好的 hook 是移除不必要的注释。agent 写代码一般都很多废话,他这个机制会让 agent 每次编辑文件都有强制精简注释的动作,很实用。

by carter003: 我觉得 OmO 还是在测试版范畴中.第一,架构非常好,通过 subagent 进行任务编排.这种架构要好于 Zen 的架构.第二,编排还不是智能的,prompt 写的比较固定,所以还不够智能.第三,由于 explore 因为网络或者项目大小.导致 background 返回可能要 1 分钟甚至更长时间.这时候主 agent 就会等不及,然后自己去探索了.相当于 explore agent 白干了.第四,偶发边界失效,比如在 spec kit 中 /plan 下一步是 /task 但是因为 OmO 遵循 Sisyphus 模式,所以自动提示大模型继续检查现有工作,这时候大模型误认为是继续执行下一步. 不过 OmO 最近好火爆.

by 沙漠白莲: 我喜欢规划一个特别特别长的规划文件,然后要 codex 运行十几个小时一次性跑完,第 2 天早上一醒来,任务就做完了 [打 call]

by 海十 Mirage: 我测试了,一个组提示词跑了一早上,然后效果并不太满意,用 cc 给同样的提示词,中途再加一些提示词,更快效果更好

插件推荐 之 bmad #

opencode manager (手机远程操作 opencode) #

手机上Coding?OpenCode Manager项目开源, (github)

opencode local+remote 本地 + 远程 mcp 配置方法 #

ref: video Bilibili by 技术爬爬虾 OpenCode详细攻略,开源版Claude Code,免费模型与神级插件

OPENCODE (VSCODE EXTENSION) #

Pre-requirement: OpenCode CLI. The doc said it can be installed automatically by running opencode in the termianl given having CLI installed. However, I need to manually install.