# 5ire: a desktop AI assistant that speaks nine providers and MCP > 5ire, pronounced fai-er, is a cross-platform desktop AI assistant and MCP client built in TypeScript on Electron, connecting to OpenAI, Azure, Anthropic, Google, Mistral, Doubao, Grok, DeepSeek and Ollama. Its tools feature runs MCP servers, its knowledge base parses and vectorizes six document formats locally with bge-m3, and its author maintains a community marketplace of MCP servers. **nanbingxyz/5ire** — 5ire is a cross-platform desktop AI assistant, MCP client. It compatible with major service providers, supports local knowledge base and tools via model context protocol servers . - Repository: https://github.com/nanbingxyz/5ire - Website: https://5ire.app - Stars: 5,365 · Forks: 433 - Language: TypeScript - License: NOASSERTION - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/nanbingxyz-5ire ## Nine providers behind one native window 5ire, which the README teaches to pronounce fai-er, is a cross-platform desktop AI assistant and MCP client, and its first promise is provider breadth, OpenAI, Azure, Anthropic, Google, Mistral, Doubao, Grok, DeepSeek and Ollama are listed side by side, mixing the American majors, a Chinese incumbent, an xAI entrant and a local runtime. That spread is the practical appeal of a dedicated desktop client over each vendor's own app, one conversation surface, one set of prompts and history, with the model underneath as a swappable setting, and the inclusion of Ollama means the same interface serves fully local models when privacy or offline work demands it. The project is listed among the clients on the Model Context Protocol's own site, which anchors its MCP credentials beyond self-description, and the author operates a Discord where feature input and support questions are directed. The website at 5ire.app carries screenshots and download links, and a DeepWiki page renders the repository's architecture for anyone evaluating the codebase before installing it, a review path that has become table stakes for open source AI tooling. ## MCP as a USB-C port, with a runtime bill For the tools feature, the README reaches for the now-standard analogy, MCP is like a USB-C port for AI applications, a standardized way to connect models to data sources and tools, so with tools enabled you can access the file system, obtain system information, interact with databases and reach remote data rather than just having a conversation. The honest part is the prerequisite list, before activating tools you need Python, Node.js and uv installed, because those constitute the runtime environment for MCP servers, and the README explicitly permits deferring the whole step until tools are actually wanted. For spreading the capability, the project publishes a one-click server installation mechanism, documented in an integration guide for embedding that install flow into other websites, so an MCP server's own site can offer add-to-5ire the way sites offer login-with. That turns the client from a consumer of a protocol into a distribution channel for it. ## MCPSvr, a marketplace grown beside the client Alongside the assistant itself, the author maintains MCPSvr, an open, community-driven directory of MCP servers, described as empowering users to discover exceptional tools while offering a streamlined process for sharing their own MCP server creations. Its existence beside the client is a small ecosystem play, the way app stores grew next to platforms, discovery is the bottleneck once a protocol standardizes, and a curated directory addresses it before the noise does. For a 5ire user the marketplace is the practical answer to the blank-slate problem every extensible client has, the moment after installation when the tools menu is empty and the documentation does not say what is worth adding, and for MCP server authors it is a publication venue with the one-click install waiting on the other side. ## A knowledge base that stays on your disk The local knowledge base integrates the bge-m3 model as its embedding engine, chosen for excelling in multilingual vectorization, and parses and vectorizes docx, xlsx, pptx, pdf, txt and csv documents, storing the resulting vectors locally to power retrieval-augmented generation. Every word in that sentence carries weight for a certain user, six formats covers the office-suite output people actually have, multilingual embeddings matter outside English-first environments, and local vector storage means the knowledge base is not a courtesy front end for a cloud embedding service with your documents transiting it. That is the differentiation a desktop assistant can offer over web chat interfaces, the same hardware that renders the UI can hold the index, and the RAG answers arrive without a second party learning what you asked about. ## The small features that decide daily use Three conveniences round out the feature list, each small and each answering a real behavior pattern. Usage analytics tracks API usage and spending, so a person juggling pay-per-token services across the nine providers can see what each conversation actually costs and adjust. The prompts library supports creating and organizing prompts with variables, turning reusable scaffolds into fill-in templates rather than copy-paste-and-edit text. Bookmarks save individual conversations independently of the underlying messages, so cleaning up history does not destroy the exchanges worth keeping, an unusually thoughtful data-lifecycle decision. Quick search performs keyword search across all conversations, the feature whose absence is felt the day a remembered answer cannot be found again. Individually none of these would sell the product, but their combined presence is what distinguishes an assistant designed for continued daily use from one assembled to demonstrate protocol support. Together these are the difference between a demo of an MCP client and a tool someone keeps open all day. ## An Electron app from the boilerplate school The repository structure tells you the lineage at a glance, an .erb directory marks the electron-react-boilerplate heritage, with the two-process dev workflow that template popularized, rsbuild serving the renderer while a nodemon-driven process restarts the main bundle, and jest configured with electron mocks for tests that run outside a live window. Modern layers sit on top, biome for linting and formatting in one tool, Tailwind for styling, husky for git hooks with a better-commits configuration for changelog-friendly messages, and, more notably, Drizzle as the ORM with its own config and migration generation, a serious choice for the local database backing conversations and the knowledge base. patch-package applies fixes to dependencies from a patches directory, and the packaging path runs through electron-builder with per-platform scripts driven by tsx. ## A modified license and the notarization toll Two adoption realities deserve plain statement. The license is one, GitHub reports no recognized license for the repository while the package manifest declares Modified Apache-2.0, a phrase that obliges anyone redistributing or embedding 5ire to read the LICENSE file and understand the modification before relying on Apache assumptions. The second is packaging, since native dependencies require the app to be packaged on the corresponding platform, and the README's tip notes that macOS builds additionally need APPLE_TEAM_ID, APPLE_ID and APPLE_ID_PASS configured for notarization to spare users security alerts, the Apple tax on shipping desktop software. Release history runs v0.15.2 in December 2025, v0.15.3 in January 2026 and v0.15.4 in March 2026, with the last push on 2026-09-14, and support flows through Discord and support@5ire.app. ## Conclusion Use 5ire when you want one native desktop application for conversational AI across many providers, including local Ollama models, with MCP tools and a genuinely local knowledge base rather than a cloud embedding service. Prefer a browser-based assistant when cross-device sync outweighs local storage, or a terminal client when your workflow lives in the shell. Verify first that your platform has a packaged build since native dependencies require per-platform packaging, read the LICENSE file before redistributing because the manifest declares a modified Apache-2.0, and install Python, Node.js and uv only when you actually enable the tools feature. ## FAQ ### What is 5ire, the AI assistant? 5ire, pronounced fai-er, is a cross-platform desktop AI assistant and MCP client. It connects to OpenAI, Azure, Anthropic, Google, Mistral, Doubao, Grok, DeepSeek and Ollama, supports tools through MCP servers, and includes a local knowledge base for retrieval-augmented generation. ### What do you need to run 5ire's tools feature? Python, Node.js and the uv package manager, which together form the runtime environment for MCP servers. The installation can be skipped initially and completed later when you actually want to enable tools. ### Which file types can 5ire's local knowledge base parse? docx, xlsx, pptx, pdf, txt and csv documents are parsed and vectorized using the bge-m3 embedding model, with the resulting vectors stored locally to power retrieval-augmented generation. ## Sources - [Issues](https://github.com/nanbingxyz/5ire/issues) - [nanbingxyz/5ire on GitHub](https://github.com/nanbingxyz/5ire) - [Project website](https://5ire.app) - [README](https://github.com/nanbingxyz/5ire/blob/main/README.md) - [Releases](https://github.com/nanbingxyz/5ire/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/nanbingxyz-5ire