v0.10.0 · OPEN SOURCE · MIT v0.10.0 · 开源项目 · MIT 协议

BITIDEA

The self-improving AI agent. Creates skills from experience, improves them during use, and runs anywhere — from a $5 VPS to a GPU cluster. 一个会自我进化的 AI 智能体。从经验中提炼技能,在使用中持续改进,可部署于任何环境 —— 小到 一台廉价云服务器,大到 GPU 集群

GET STARTED → 立即开始 → VIEW SOURCE 查看源码
~/bitidea · install.sh
$ curl -fsSL https://raw.githubusercontent.com/sabrinaberger083-sys/bitidea-agent/main/scripts/install.sh | bash
200+
LLM Models大模型
50+
Built-in Tools内置工具
70+
Skills智能体技能
6
Platforms消息平台

Built for autonomous intelligence. 自主智能 而生。

Not a chatbot. An agent with a closed learning loop — it curates its own memory, creates its own skills, and searches its own past. 它不是聊天机器人,而是一个拥有闭环学习能力的智能体 —— 会整理自己的记忆、创造自己的技能、检索自己过往的对话。

01
LEARN_LOOP
Self-improving memory 自我进化的记忆
Agent-curated memory with periodic nudges. Skills improve during use. FTS5 session search with LLM summarization for cross-session recall. 由智能体自己管理的记忆,带定期触发机制。技能在使用中自我打磨。基于 FTS5 全文检索 + 大模型总结,实现跨会话记忆召回。
02
OMNI_CHANNEL
Lives where you do 无处不在
Telegram, Discord, Slack, WhatsApp, Signal, and CLI — all from a single gateway process. Voice memo transcription. Cross-platform continuity. Telegram、Discord、Slack、WhatsApp、Signal 以及命令行 —— 全部通过单一网关进程接入。支持语音消息转录,跨平台对话无缝衔接。
03
MODEL_AGNOSTIC
Any model, any provider 任意模型,任意厂商
OpenRouter, OpenAI, Anthropic, Xiaomi MiMo, z.ai, Kimi, MiniMax, Hugging Face, or your own endpoint. Switch with one command — no code changes. OpenRouter、OpenAI、Anthropic、小米 MiMo、z.ai、Kimi、MiniMax、Hugging Face,或你自己的接口。一行命令切换 —— 无需改代码,无厂商锁定。
04
CRON_AUTO
Scheduled autonomy 定时自主任务
Built-in cron scheduler with delivery to any platform. Daily reports, nightly backups, weekly audits — all in natural language, running unattended. 内置 cron 调度器,可分发到任意消息平台。每日汇报、夜间备份、周审计 —— 全部用自然语言描述,无人值守自动运行。
05
PARALLEL_EXEC
Delegates & parallelizes 任务分派与并行
Spawn isolated subagents for parallel workstreams. Python scripts call tools via RPC, collapsing multi-step pipelines into zero-context-cost turns. 可派生隔离的子智能体并行处理任务。Python 脚本通过 RPC 调用工具,把多步流程压缩为零上下文成本的单次对话。
06
ANY_RUNTIME
Runs anywhere 随处运行
Six terminal backends: local, Docker, SSH, Daytona, Singularity, Modal. Serverless hibernation — costs nearly nothing between sessions. 六种终端后端:本地、Docker、SSH、Daytona、Singularity、Modal。支持 Serverless 休眠 —— 会话之间的空闲成本几乎为零。

Tools are blocks. Skills are blueprints. 工具是 积木,技能是 图纸

Two layers working together. Tools give the agent atomic capabilities. Skills teach it how to combine them — and it can write new skills on its own. 两层协作架构。工具提供原子能力,技能教它如何组合这些能力 —— 而且它会自己写新的技能。

LAYER_01 · TOOLS 50+
Tools 工具 / Tools
// atomic capabilities · hard-coded // 原子能力 · 代码实现
Python functions the agent can call: web search, file I/O, shell execution, vision, TTS. Each tool does one thing, exposed via a clean schema. 智能体可以直接调用的 Python 函数:网页搜索、文件读写、命令执行、图像识别、语音合成。每个工具职责单一,通过清晰的 schema 暴露。
×
LAYER_02 · SKILLS 70+
Skills 技能 / Skills
// workflows · natural language // 工作流 · 自然语言
Markdown files describing HOW to combine tools for a task. Self-created from experience, self-improved during use. Compatible with the agentskills.io open standard. 用 Markdown 描述的工作流,告诉智能体如何组合工具完成任务。从经验中自动生成,使用中自我优化。兼容 agentskills.io 开放标准。
01 Web & Browser网络与浏览器 4
web_tools browser_tool browser_camofox url_safety
02 Files & Code文件与代码 5
file_operations file_tools code_execution_tool terminal_tool patch_parser
03 Vision & Media视觉与多媒体 5
vision_tools image_generation_tool tts_tool transcription_tools voice_mode
04 Memory & Search记忆与检索 6
memory_tool session_search_tool skill_manager_tool skills_tool skills_hub skills_sync
05 Orchestration任务编排 6
delegate_tool mixture_of_agents_tool cronjob_tools todo_tool checkpoint_manager clarify_tool
06 External Integrations外部集成 5
send_message_tool homeassistant_tool mcp_tool mcp_oauth rl_training_tool

// + 24 internal infrastructure modules (security, approval, registry, credential handling, etc.) // + 24 个底层基础设施模块(安全、审批、注册、凭证管理等)

01 MLOps机器学习运维 21
modal lm-evaluation-harness weights-and-biases huggingface-hub gguf guidance llama-cpp obliteratus outlines vllm audiocraft clip segment-anything stable-diffusion whisper dspy axolotl grpo-rl-training peft pytorch-fsdp trl-fine-tuning unsloth
02 Creative创意 9
architecture-diagram ascii-art ascii-video creative-ideation excalidraw manim-video p5js popular-web-designs songwriting-and-ai-music
03 GitHubGitHub 工作流 6
codebase-inspection github-auth github-code-review github-issues github-pr-workflow github-repo-management
04 Productivity效率办公 6
google-workspace linear nano-pdf notion ocr-and-documents powerpoint
05 Software Development软件开发 6
plan requesting-code-review subagent-driven-development systematic-debugging test-driven-development writing-plans
06 Research研究分析 5
arxiv blogwatcher llm-wiki polymarket research-paper-writing
07 Apple Ecosystem苹果生态 4
apple-notes apple-reminders findmy imessage
08 Autonomous Agents自主智能体 4
bitidea-agent claude-code codex opencode
09 Media媒体内容 4
gif-search heartmula songsee youtube-content
10 Gaming游戏 2
minecraft-modpack-server pokemon-player
11 MCP IntegrationMCP 集成 2
mcporter native-mcp
12 Other Domains其他领域 10
jupyter-live-kernel · data-science webhook-subscriptions · devops himalaya · email find-nearby · leisure obsidian · note-taking godmode · red-teaming openhue · smart-home xitter · social-media dogfood

// Skills install via bitidea skills install <name> · write your own with a single SKILL.md file // 通过 bitidea skills install <名称> 安装 · 也可以用一个 SKILL.md 文件定义自己的技能

// SELF_IMPROVING_LOOP // 自我进化闭环
Use tools → extract pattern → save as skill → reuse → refine 调用工具 → 识别模式 → 沉淀为技能 → 复用 → 持续打磨
CALL调用 OBSERVE观察 ABSTRACT抽象 PERSIST固化 REFINE精进
The agent doesn't just execute tasks — it learns from them. Every non-trivial workflow becomes a skill for next time. 智能体不只是执行任务 —— 它还从任务中学习。每一次复杂的工作流,都会沉淀为下次可复用的技能。

One command. Zero friction. 一行命令,零配置启动。

Works on Linux, macOS, WSL2, and Android via Termux. Installs the CLI, sets up config, and gets you chatting in under 2 minutes. 支持 Linux、macOS、WSL2,以及 Android(通过 Termux)。自动安装命令行工具、配置环境,两分钟内即可对话。

~/bitidea · bash
# 1. install $ curl -fsSL https://raw.githubusercontent.com/sabrinaberger083-sys/bitidea-agent/main/scripts/install.sh | bash # 2. reload shell and start $ source ~/.bashrc # or: source ~/.zshrc $ bitidea # 3. choose your model $ bitidea model # 4. start messaging gateway (optional) $ bitidea gateway
# 1. 安装 $ curl -fsSL https://raw.githubusercontent.com/sabrinaberger083-sys/bitidea-agent/main/scripts/install.sh | bash # 2. 重载 shell 并启动 $ source ~/.bashrc # 或: source ~/.zshrc $ bitidea # 3. 选择你的模型 $ bitidea model # 4. 启动消息网关(可选) $ bitidea gateway

Plugs into your stack. 接入你的 技术栈

Any LLM provider. Any messaging platform. Any runtime. No vendor lock-in, no hidden dependencies. 任意大模型厂商、任意消息平台、任意运行时。没有厂商锁定,没有隐性依赖。

OpenRouter
OpenAI
Anthropic
z.ai / GLM
Kimi / Moonshot
Xiaomi MiMo小米 MiMo
MiniMax
Hugging Face
Mistral
AWS Bedrock
Telegram
Discord
Slack
WhatsApp
Signal
Matrix
Feishu飞书
DingTalk钉钉
Home Assistant
Email / SMS邮件 / 短信
Local本地
Docker
SSH
Daytona
Singularity
Modal
MCP Servers
Skills Hub
Honcho
Atropos RL

Your agent. Your rules. 你的智能体,由你掌控。

Open source. MIT licensed. Run it on your laptop, your VPS, or your GPU cluster. Talk to it from anywhere. 开源项目,MIT 协议。部署在笔记本、VPS 或 GPU 集群上都可以 —— 从任何地方与它对话。