Container v0.1 一款 免费Local-first 的大模型桌面工作台：集成大模型部署、管理、对话、蜂群、工具调用（MCP）、RAG 知识库、agent skill、桌面自动化操作与 OpenAI 兼容 API 网关，让本地推理真正可用、可控、可扩展。 把大模型从“一个人聊天”升级为“一个团队工作”： 蜂群多智能体编排 + 工具系统 + 本地推理与 API 网关，一体化交付可执行结果。

Highlights

• 开箱即用的桌面体验： Local-first 推理，模型与数据主要在本机处理，延迟低、可控性强聊天（可选免费api，无需部署） + 参数面板 + 控制台 + 资源监控，所有关键状态可视化
• OpenAI 兼容 API 网关：本地应用也能被你的工具链直接调用（SSE 流式、模型列表、embeddings 等）
• MCP 工具生态：一键安装自定义工具，基本预设满足日常工作需求
• RAG 知识库：目录/单文件索引、刷新与清理索引、可控召回数量与阈值
• 工作区与检索：文件树、文本搜索、语义搜索（依赖支持 indexing/search 的 MCP 服务）
• 运行时引擎管理：管理本地推理引擎（Windows：CUDA / Vulkan / CPU），支持下载、切换版本与额外启动参数 Core Features
• 模型管理 支持本地模型导入/删除 集成模型搜索与下载（支持展示模型信息）
• 对话与推理控制 温度、TopP/TopK、重复惩罚、上下文长度、batch/线程等参数可视化智能配置 GPU Offload（含自动/手动策略）、显存压力提示与实际加载层数展示 Structured Output：支持 JSON Schema 约束输出
• 多模态与工具 图片理解：可配置视觉模型 工具调用：支持在对话中手动运行 / 自动触发 MCP 工具
• 语音输入 支持本地/云端语音转文字（可控配置本地 ASR 引擎）
• 网络与安全 网关支持：端口、CORS、API Key 支持上游转发（Upstream Base URL + Key）与网络代理
• 系统可观测性 底栏实时展示 CPU / RAM / GPU / VRAM 等指标（不同硬件/驱动能力下可能有所差异）
• Swarm 蜂群多智能体 角色编排、层级关系、可视化、导入导出、角色预算控制
• 工作区与检索
• • 文件树、文本搜索、语义搜索，终端自动化创建项目
• 并发测试
• • 测试模型能力，优化调参
• 隐私优先（Local-first）
• -你的对话、提示词、知识库与模型文件默认只保留在本机，我们不做内容采集与用户画像；为保证稳定性，仅在发生异常时上传 最小化 诊断信息用于修复（错误级别/摘要、堆栈、页面路由、客户端版本与基础设备信息），不包含对话内容、文件内容或密钥等私人信息。 Quick Start 下载并安装：见 Release Assets ��开应用 → 设置模型目录与运行时目录（可选） 在 Models 中导入/下载模型 进入 Chat 页面加载模型并开始对话 如需 API：打开 Server 页面启用网关并配置端口/API Key Notes 视觉理解、语音、本地 GPU 指标采集等能力受系统环境与驱动影响；建议保持显卡驱动更新。 语义搜索需要已安装并启用支持 indexing/search 的 MCP 服务，并先建立索引。

Changelog V0.1 Full Changelog: https://github.com/peoples111/container/commits/llm

Container v0.1 A free Local-first Large Language Model Desktop Workbench: Integrating model management, chat, swarm orchestration, tool calling (MCP), RAG knowledge bases, workspace automation, and an OpenAI-compatible API gateway. We make local inference truly usable, controllable, and scalable. Upgrade your workflow from "chatting with one AI" to "working with a team." Container delivers an all-in-one solution combining Swarm multi-agent orchestration, a robust tool system, and local inference/API gateway capabilities to produce executable results. 🌟 Highlights Out-of-the-Box Desktop Experience: Local-first Inference: Models and data are processed primarily on your machine for low latency and maximum control. Flexible Connectivity: Chat using local models or optional free cloud APIs (no deployment required). Full Visibility: Visualize everything with parameter panels, console logs, and real-time resource monitoring. OpenAI-Compatible API Gateway: Seamlessly integrate local models into your existing toolchains. Supports SSE streaming, model listing, embeddings, and more. MCP Tool Ecosystem: One-click installation for custom tools. Comes with pre-configured tools to meet daily workflow needs. RAG Knowledge Base: Index directories or single files. Full control over index refreshing, cleanup, recall limits, and similarity thresholds. Workspace & Retrieval: Integrated file tree, text search, and semantic search (powered by MCP services with indexing/search capabilities). Runtime Engine Management: Manage local inference engines seamlessly. Windows Support: CUDA, Vulkan, and CPU backends. Features include version switching, download management, and custom startup arguments. 🚀 Core Features Model Management Import/Delete: Easily manage local model files. Discovery: Integrated model search and download with detailed model information cards. Chat & Inference Control Visual Parameter Tuning: Intuitive controls for Temperature, Top-P/Top-K, Repetition Penalty, Context Length, Batch Size, and Threads. GPU Offloading: Smart auto/manual strategies with VRAM pressure warnings and real-time layer loading visualization. Structured Output: Enforce JSON Schema constraints for reliable programmatic responses. Multimodal & Tools Vision: Configurable visual models for image understanding. Tool Calling: Manually trigger or auto-execute MCP tools directly within the chat flow. Voice Input Flexible ASR: Support for both local and cloud-based Speech-to-Text, with configurable local ASR engine options. Network & Security Gateway Config: Customizable ports, CORS policies, and API Key management. Upstream Forwarding: Configure upstream base URLs and keys for proxying requests; supports network proxies. System Observability Real-time Metrics: Live dashboard in the footer displaying CPU, RAM, GPU, and VRAM usage (capabilities depend on hardware/drivers). 🐝 Swarm Multi-Agent Orchestration Role编排 (Role Orchestration): Define agent roles, hierarchical relationships, and interaction flows. Visualization: Visual graph of agent interactions. Management: Import/export configurations and set role-specific budget/token limits. Workspace & Automation File Operations: File tree navigation, text search, and semantic search. Terminal Automation: Automatically scaffold and create projects via terminal commands executed by agents. Concurrency Testing Built-in tools to stress-test model capabilities and optimize hyperparameters under load. 🔒 Privacy First (Local-first) Your Data Stays Yours: Conversations, prompts, knowledge bases, and model files remain strictly on your device by default. We do not collect content or build user profiles. Minimal Diagnostics: To ensure stability, only minimal diagnostic data (error level/summary, stack traces, page route, client version, and basic device info) is uploaded upon crash. No conversation content, file data, or secrets are ever transmitted. ⚡ Quick Start Download & Install: Grab the installer from the Release Assets. Setup: Launch the app → Set your Model Directory and Runtime Directory (optional). Get Models: Go to the Models tab to import existing models or download new ones. Start Chatting: Navigate to the Chat page, load a model, and begin. Enable API: Need an API? Go to the Server page, enable the gateway, and configure your Port/API Key. 📝 Notes Hardware Dependencies: Capabilities like Vision, Voice, and local GPU metrics depend on your system environment and drivers. Please ensure your graphics drivers are up to date. Semantic Search Requirement: Semantic search requires an installed and enabled MCP service that supports indexing and search. You must build the index before searching. 📅 Changelog v0.1: Initial Release.