Cua

CowAgent is an open-source AI assistant and agent harness developed by LinkAI. It plans complex tasks step by step, runs tools and skills, controls computers and external services, maintains long-term memory and a personal knowledge base, and supports multi-agent teams with separate roles, models, skills, knowledge, and workspaces. Its architecture routes messages from web and messaging channels through an agent core that reasons over available models, tools, skills, memory, and knowledge before returning a response. Memory uses a three-tier structure of conversation context, daily memory, and long-term MEMORY.md, with a nightly Deep Dream process that distills memories; the knowledge base is an automatically curated Markdown wiki with cross-references and a browsable knowledge graph. Built-in tools include file and terminal operations, web fetching and search, scheduling, memory retrieval, vision, and browser automation, while MCP servers can be added through configuration. Skills can be installed from the Skill Hub, GitHub, ClawHub, or URLs, or authored through conversation. CowAgent supports multiple LLM providers and routes chat, vision, image generation, speech, and embedding workloads separately. It can run locally, in Docker, on a server, or through its macOS and Windows desktop client, with a web console for configuration and a command-line interface for service and skill management. The project is licensed under the MIT License and was formerly named chatgpt-on-wechat.

Munder Difflin is a free, open-source desktop multi-agent harness that runs multiple terminal-based AI coding agents as coordinated local agents. It wraps agents including Claude Code, OpenAI Codex, Gemini CLI, Qwen, OpenCode, GitHub Copilot CLI, and custom commands, working with users' existing subscriptions and hourly usage limits. Each agent runs as a real node-pty process and is rendered through xterm.js, while an Electron and React interface displays sessions as avatars on a Pixi.js office floor. A GOD orchestrator called Michael assigns and routes work, adjudicates agent messages, and escalates spending, destructive operations, scope changes, and other configured approvals. Agents coordinate through a local Git-backed hive of plain-file memory, atomic mailboxes, a shared blackboard, and an append-only event log; a markdown-first semantic memory layer supports recall across sessions. Optional Git worktrees isolate parallel agents.

U-Pool is a desktop provider and account switcher for Claude Code, Claude Desktop, Codex, Hermes, OpenCode, and Cursor. It keeps endpoint URLs, API keys, models, and provider accounts in per-tool lists, then makes a selected provider or Cursor account active without manual edits to settings.json, config.toml, other tool configuration files, or Windows user environment variables. Its reachability probe sends a plain GET request to a provider’s models endpoint to measure latency without running a completion or consuming a token. The app uses a Python backend, a statically exported Next.js interface, and a native OS webview rather than Electron or a Node runtime. Its source of truth is ~/.u-pool/config.json; on each switch, an adapter changes only the keys it owns in each target file and preserves other settings such as plugins, hooks, marketplaces, themes, MCP servers, and project trust. Writes use a temporary file followed by os.replace, keep an adjacent backup, and stop rather than overwrite a configuration file that cannot be parsed. Existing provider settings can be imported on first run. The app includes provider presets, an official-vendor mode that clears U-Pool-managed settings, per-provider permission controls, cached installed Claude Code and Codex versions, and controls that erase the transcript and prompt-history files for those CLIs. The Cursor pool stores multiple signed-in Cursor accounts, accepts cookies in common formats, displays account names, email addresses, plans, and usage, and switches accounts by changing the specified Cursor authentication database rows. Claude Desktop is currently preview-only: providers can be saved and tested but are not written to its configuration. Hermes and OpenCode have dedicated configuration writers, while Codex writes its provider, model, approval, sandbox, web-search, and authentication settings. The core configuration operations are cross-platform. Windows-only integrations write owned variables to HKCU\\Environment and broadcast changes, launch the app at sign-in, and provide an in-app updater; Hermes, OpenCode, and Cursor read credentials from their own files and do not use those environment-variable writes.

Mu is a self-hosted personal-agent platform that provides one home for agents, tools, services, and collected data. Its server can be accessed through a web app, email, MCP, HTTP, and a command-line interface, and supports bringing your own model. Mu exposes services for mail, chat, news, video, markets, search, files, calendars, and small web applications. These services are available through APIs, the CLI, the web, or as MCP tools. Data collected by the services is archived locally so it can be searched across sources. Users can define agents by name, prompt, and permitted tools, then chat with them through the web, mail, or XMPP. The platform also provides a unified inbox for email and chat threads, including access through IMAP.