Local-first operation
Core inference and conversation history stay on the machine after required local components are installed.
Loading OpenMindAI
Loading OpenMindAI
OpenMindAI is a local-first desktop & Mobile AI workspace built around privacy, portable storage and practical offline intelligence.
The project exists because the most important part of a desktop AI workflow—inference and working context—should not require a remote service by default.
Core inference and conversation history stay on the machine after required local components are installed.
Models and application data can live on a secondary drive, external SSD or portable installation root.
The workstation detects host hardware and selects an appropriate local inference path instead of assuming one GPU profile.
Chat, models, projects, files, tools, diagnostics, updates and maintenance belong in the same product.
OpenMindAI uses React + TypeScript + Vite inside a Tauri 2 desktop shell. Native application responsibilities live in Rust, while SQLite and local runtime/model files stay in user-controlled storage.
Downloaded models, runtimes, local databases, caches, logs, generated content, workspaces and backups stay outside Git. Large GGUF and SafeTensors files must not enter normal repository history.
CI validates the frontend and Rust code paths, while separate security and tagged-release workflows enforce dependency scans, CodeQL, checksums and signing gates.
Explore the repositoryOpenMindAI is a local-first artificial intelligence platform built for people who want powerful AI without giving up control of their devices, files, workflows, and data. It brings together private AI chat, reasoning, research, document understanding, vision, voice, local model management, project workspaces, and connected tools in one unified experience across Windows, macOS, Linux, Android, and iOS.
The idea behind OpenMindAI is simple: everyday AI should not always require a cloud subscription, a permanent internet connection, or a remote server. Whenever the required local model and runtime are available, OpenMindAI is designed to perform as much work as possible directly on the user's own device. When an online capability is genuinely useful, such as web search, deep research, software updates, model downloads, or connected services, OpenMindAI can use the internet intentionally rather than making every interaction dependent on it.
OpenMindAI is therefore not just an offline chatbot. It is a growing personal AI environment designed for real work on desktop and mobile.
Our mission is to make capable artificial intelligence more private, accessible, practical, and user-controlled.
OpenMindAI is designed to give people a way to use modern AI on their own hardware, work with their own files, select models that match their device and needs, and continue using essential AI features even when cloud services are unavailable. We want local AI to feel approachable enough for everyday users while remaining powerful enough for developers, researchers, engineers, students, and technical professionals.
The project also aims to reduce unnecessary dependence on paid AI APIs for ordinary personal workflows. Instead of forcing every request through an external provider, OpenMindAI gives users the option to run compatible models locally and decide when online services should be involved.
We believe the future of personal AI will be hybrid.
Some tasks are best handled locally because privacy, speed, availability, or data ownership matters. Other tasks benefit from live internet access, connected applications, or external information. OpenMindAI is being built around that reality.
Our vision is an AI workspace where a person can move naturally between private local intelligence and optional connected intelligence without changing tools. A user should be able to chat offline, analyze a private document locally, open a project workspace on desktop, use a local coding agent, ask a phone camera about something in front of them, and then switch to online Search or Research only when fresh public information is actually needed.
The long-term goal is to make OpenMindAI a personal AI environment that belongs to the user rather than an experience that exists only inside someone else's cloud.
Many AI products are designed around a cloud-first model. That approach can be convenient, but it also means the user may depend on internet availability, account restrictions, usage limits, subscription tiers, remote processing, and a model chosen entirely by the service provider.
OpenMindAI takes a different approach. It is designed to let the user decide where AI runs, which compatible models are installed, where local data is stored, and when online capabilities are used. This makes the platform especially useful for people who care about privacy, offline access, local development, model flexibility, or working with sensitive local files.
OpenMindAI also treats different devices differently. A desktop workstation and a mobile phone do not have the same memory, thermal limits, storage, or GPU capability. Rather than pretending that one model fits every device, OpenMindAI is designed around hardware-aware model selection and task routing. Smaller models can provide fast everyday assistance, while larger or specialized models can be used when the hardware and task justify them.
| Area | OpenMindAI | Typical Cloud-Only AI |
|---|---|---|
| AI execution | Local-first, with online capabilities when needed | Primarily remote server processing |
| Internet requirement | Many core workflows can continue locally after setup | Usually required for every interaction |
| Model control | Compatible local models can be installed and managed by the user | Model choice is controlled mainly by the provider |
| Data location | Local conversations, models, project data, and settings can remain on the device | Prompts and processing normally depend on remote infrastructure |
| Paid API requirement | Not required for normal local AI workflows | Often tied to subscription or provider usage |
| Hardware use | Uses the user's own compatible CPU/GPU resources | Processing happens mainly in provider infrastructure |
| Offline work | Supported for many local features | Usually unavailable |
| Connected information | Used intentionally for Search, Research, updates, and integrations | Connectivity is part of the core experience |
| Desktop project work | Local files, projects, workspaces, and controlled agents | Depends on the product and cloud permissions |
| Mobile AI | Designed to support on-device models and local features | Frequently relies on remote inference |
OpenMindAI is designed to fit into real everyday work instead of existing only as a question-and-answer window.
For a student, it can explain difficult subjects, summarize learning material, organize notes, analyze documents, and help build a study plan. For a researcher, it can work with papers and technical documents, assist with literature exploration, compare approaches, help structure research ideas, and use online research modes when current public evidence is needed.
For a developer, OpenMindAI can become part of the local development workflow. It can explain code, review project files, help debug errors, prepare documentation, compare architectures, inspect source context, and support project-based agent workflows. On desktop, project workspaces are intended to keep instructions, conversations, reference files, source code, and agent context together so the AI can work within a defined scope rather than starting from zero in every chat.
For writers and professionals, OpenMindAI can help draft reports, emails, documentation, articles, proposals, summaries, technical descriptions, and structured documents. Because many of these workflows can run locally, users can work with private material without automatically sending every prompt to a remote AI service.
For everyday mobile use, OpenMindAI can provide a portable assistant that travels with the user. A person can speak instead of typing, attach a document, use the camera as visual context, ask questions about an image, continue a conversation, or use a locally installed model even when internet access is poor.
OpenMindAI is designed around several common ways of working.
A user can begin with a normal conversation and use the locally installed model for questions, writing, planning, brainstorming, or technical help. When a problem needs deeper reasoning, a reasoning-oriented mode or compatible model can be used. When current information is required, Search or Research can retrieve public web evidence and give that context to the AI for analysis.
Documents and files can be added when the user wants the AI to work with specific material. Depending on platform and supported formats, this may include PDF, DOCX, Markdown, text files, source code, JSON, YAML, CSV, and other text-based content. Instead of copying large amounts of material manually into a chat, the user can work directly from the relevant file context.
On desktop, users can move beyond isolated chats by creating project-oriented workspaces. These can hold project instructions, files, references, conversations, and agent context. A controlled local agent can then help inspect, create, modify, or validate files inside the approved project scope.
On mobile, the experience is optimized for portability. Users can download suitable on-device models, use chat, thinking, vision, voice, document input, local conversation history, Search, Research, and other mobile-focused features from Android or iOS.
OpenMindAI is intentionally transparent about the difference between local capabilities and online capabilities.
The core philosophy is to keep work local whenever practical, while still allowing the user to benefit from the internet when a feature genuinely depends on live information or a remote service.
| Can Work Locally After Setup | Normally Requires Internet |
|---|---|
| Local AI chat | Initial model downloads |
| Local reasoning | Runtime downloads |
| Conversation history | Software updates |
| Local document analysis | Web Search |
| Project files and local context | Deep Research evidence retrieval |
| Local model execution | Connected applications |
| Supported local vision | GitHub remote access |
| Supported local voice features | Gmail, Google Workspace, Microsoft 365, Slack, Notion, Dropbox and similar services |
| Local project workflows | Other remote APIs or services |
| Local Canvas workflows | Any feature that depends on live public information |
This hybrid approach is important because it avoids two extremes. OpenMindAI does not pretend that everything useful can happen without a network, and it also does not force every AI interaction to leave the device.
OpenMindAI is designed as one ecosystem, but desktop and mobile are optimized for different types of work.
| Platform | Best Suited For |
|---|---|
| Windows | Full desktop AI workspace, local models, documents, projects, agents, development workflows, connected apps, hardware-aware execution |
| macOS | Local AI productivity, research, development, documents, projects, supported hardware acceleration and connected workflows |
| Linux | Developer and research workflows, local models, project workspaces, source analysis, technical automation and open computing environments |
| Android | Portable local chat, model downloads, voice, camera input, vision, documents, Search, Research and everyday on-device assistance |
| iOS | On-device AI, chat, reasoning, voice, vision, document workflows, local history, mobile research and supported local generation features |
Desktop systems generally have more memory, storage, cooling, and computing capability, which makes them better suited for larger models, long-running workloads, project agents, and deeper file-based work. Mobile devices are more constrained, so OpenMindAI mobile is designed to balance model quality with memory, battery, storage, and thermal limits.
Different tasks benefit from different models.
A fast lightweight model may be ideal for short conversations on a phone. A reasoning model may be more useful for complex technical analysis. A vision-capable model is required when the user wants the AI to understand an image. A larger desktop model may produce better results but may require significantly more memory and processing power.
OpenMindAI is designed to take those differences into account.
Its model-routing approach can consider the kind of task being performed, whether vision is required, which compatible models are installed, available memory, device architecture, runtime compatibility, and user preferences. This makes it possible for OpenMindAI to provide a more practical experience across very different hardware instead of assuming that every user owns a high-end AI workstation.
OpenMindAI gives users control over the models they keep on their devices.
Models can be downloaded separately, allowing the application itself to remain smaller and letting users decide which capabilities they actually need. A user who only needs lightweight chat can keep a smaller model, while someone who needs stronger reasoning, vision, or desktop performance can install additional compatible models.
The model management experience is intended to make large local AI files easier to handle by supporting installation status, download progress, resumable downloads where supported, file verification, compatibility checks, model removal, and storage management.
This is particularly important on mobile, where AI model files can take hundreds of megabytes or several gigabytes and storage should remain under the user's control.
Privacy is not treated as a single setting inside OpenMindAI. It is part of the architecture.
When a supported feature runs locally, the model can process prompts and local context directly on the user's device. Conversation history, settings, models, project context, and other application data can remain in local storage rather than being automatically routed through a third-party AI provider.
That does not mean OpenMindAI never communicates with online services. If the user starts a web search, downloads a model, checks for an update, or connects an external account, network access is obviously required. The difference is that those online actions are tied to capabilities the user intentionally chooses.
OpenMindAI's goal is to make the boundary understandable: local work should remain local where possible, and connected work should be used when it adds value.
A large part of useful AI work begins with existing information rather than an empty chat box.
OpenMindAI is designed to work with local documents and files so the AI can help explain, summarize, compare, rewrite, inspect, or organize material that already exists. This can include reports, academic papers, notes, technical documentation, source code, structured data, and other supported file types.
For a researcher, that might mean reviewing a paper and identifying its key claims. For a developer, it might mean asking about a source file or configuration. For a professional, it could mean turning an internal draft into a clearer document. For a student, it could mean converting a long PDF into understandable study notes.
The purpose is to move AI from generic conversation into context-aware work.
On desktop, OpenMindAI Projects are intended for work that continues over time.
A project can bring together instructions, conversations, reference material, attached folders, source code, and task context. This creates a stable environment for long-running work such as software development, research, documentation, or technical planning.
A local project agent can then operate within the approved workspace. Depending on the permissions granted by the user, the agent may inspect files, read source material, create or edit files, review project structure, validate changes, inspect Git state, or help repair failures.
The key principle is control. Local agents should not automatically receive unlimited access to the entire computer. Broader file-system or terminal capabilities remain explicit permission boundaries so the user understands what the agent is allowed to do.
Local models are useful, but they do not automatically know what happened today.
OpenMindAI addresses that limitation by separating local reasoning from live information retrieval.
Search can retrieve current public web information and pass that evidence into the AI workflow. Research can go further by gathering and organizing information from multiple public sources so the AI has stronger context for a detailed answer.
This allows users to keep local AI as the reasoning engine while still accessing current information when necessary. It also means OpenMindAI does not need to send every normal conversation to the internet simply because some tasks occasionally require fresh data.
OpenMindAI supports multimodal workflows when a compatible local vision model is available.
An image can become part of the conversation just like text. Users may ask about a screenshot, diagram, document image, interface, photograph, or other visual material. On mobile, the camera can provide immediate visual context, making the AI useful in situations where typing a description would be slow or inaccurate.
The value of vision is not simply image recognition. It gives the AI additional context so it can explain what is visible, reason about a screenshot, help understand a diagram, or answer questions about something the user has photographed.
OpenMindAI is designed to support both speech input and spoken responses.
Voice input allows users to dictate prompts, capture ideas, take notes, or interact without typing. Text-to-speech can read responses aloud, which can improve accessibility and make mobile or hands-free use more practical.
These capabilities help OpenMindAI feel less like a traditional desktop utility and more like a personal assistant that can adapt to different ways of interacting.
Local-first does not mean isolated.
OpenMindAI can optionally connect with external services such as Google Workspace, GitHub, Microsoft 365, Slack, Notion, Dropbox, and compatible tool ecosystems where supported. These integrations are intended for users who want their AI workspace to interact with the services they already use.
A developer might connect GitHub to work with repositories and development activity. A professional might connect a productivity service. A team might connect communication or document tools.
These capabilities are optional and depend on user authorization. Credentials and authentication data should remain separate from normal chat content, and remote actions should stay subject to the permissions granted by the user.
OpenMindAI can act as a private study companion. It can explain complex ideas in simpler language, summarize documents, help organize notes, generate revision material, compare concepts, assist with research, and support learning even when internet access is unavailable.
Researchers can use OpenMindAI to work with technical material, analyze documents, organize ideas, compare methods, prepare drafts, explore public sources, and keep much of the research workflow local. Search and Research can be used when current evidence is required.
Developers can use OpenMindAI to understand code, debug errors, review architecture, generate documentation, explore project files, work with Git context, compare implementation options, and use controlled local agents inside project workspaces.
OpenMindAI can help with technical reasoning, system architecture, troubleshooting, documentation, research, requirement analysis, and structured planning without forcing sensitive internal material into a cloud AI workflow.
Writers, founders, managers, consultants, and other professionals can use OpenMindAI for reports, proposals, articles, emails, summaries, planning, brainstorming, and document preparation while keeping more working context on their own devices.
People who simply want more control over AI can use OpenMindAI without making every conversation dependent on a remote commercial AI service.
Getting started is designed to be straightforward.
On desktop, the user installs OpenMindAI, selects or confirms the location where application data and models will be stored, allows the application to detect the available hardware, and installs a compatible local AI model. Once that model is ready, normal local chat can begin.
Additional models can then be installed when the user wants stronger reasoning, different capabilities, or support for a specialized task. Documents can be attached directly to conversations, while Projects can be used for longer work involving folders, code, files, or local agents.
On mobile, the application guides the user through onboarding and model installation. The user can begin with a model suitable for the device and later add other compatible models depending on storage, memory, and performance requirements.
Internet access is mainly needed during setup for model or runtime downloads and later whenever the user intentionally chooses an online capability such as Search, Research, updates, or connected services.
OpenMindAI is not trying to make a phone behave exactly like a workstation.
The desktop experience is designed for larger models, longer sessions, local projects, development, document generation, connected tools, and deeper agent-assisted work. The mobile experience is designed for portability, fast access, voice, camera, document input, local models, and everyday AI assistance.
Both applications share the same core idea: the device should be able to provide useful AI directly, while online services remain available when they are actually needed.
This allows a user to think of OpenMindAI as one AI ecosystem rather than separate products with unrelated purposes.
Running AI locally gives users more control, but it also creates new responsibilities.
OpenMindAI is designed around permission boundaries rather than assuming that an AI model should automatically be trusted with unrestricted access. Project agents, file operations, connected services, and terminal-level actions should remain controlled and scoped.
The platform also treats external web content as information rather than trusted instructions. This distinction is important when Search or Research brings public internet content into an AI workflow.
Downloaded models and runtime components may also be verified when the necessary metadata is available, helping reduce the risk of incomplete or corrupted files being used.
Local AI does not have one fixed performance level.
The experience depends on the CPU, GPU, available memory, model size, quantization, context length, operating system, runtime, storage speed, and, on mobile, battery and thermal conditions.
A small model may respond quickly on modest hardware and be perfect for everyday use. A larger model may provide stronger answers but need much more RAM and processing power. A vision model adds additional requirements. Mobile devices require even more careful balancing because they operate within tighter memory and power limits.
OpenMindAI's hardware-aware design exists to make those trade-offs easier for the user rather than hiding them.
OpenMindAI is not intended to claim that local AI removes every limitation of artificial intelligence.
Local models can still make mistakes, misunderstand context, produce incomplete information, or generate incorrect answers. Important medical, legal, financial, scientific, security, and safety-critical information should always be verified with appropriate trusted sources or qualified professionals.
OpenMindAI is also not an offline-only product. Web Search, Research, model downloads, updates, and connected applications naturally require a network connection. The goal is not to eliminate the internet; it is to make the internet optional for workflows that do not genuinely need it.
OpenMindAI is built around three ideas: local when possible, connected when useful, and controlled by the user.
Local when possible means everyday AI should be able to run on the hardware people already own.
Connected when useful means users should still be able to access current information, updates, and external services when those capabilities improve the task.
Controlled by the user means models, data, storage, project permissions, and connected actions should remain understandable and manageable rather than hidden behind a cloud service.
That philosophy is the foundation of the entire OpenMindAI ecosystem.
OpenMindAI is being developed toward a future in which personal AI becomes a standard part of the operating environment rather than only a website or subscription service.
As local models become more efficient and consumer hardware becomes more capable, OpenMindAI aims to make those improvements directly useful to everyday users. The platform will continue evolving around better local inference, stronger model routing, richer multimodal interaction, safer local agents, improved project workflows, deeper document intelligence, and seamless movement between local and connected capabilities.
The goal is not simply to create another AI chat application.
The goal is to build an AI environment that users can actually own, understand, customize, and rely on.
OpenMindAI brings private local intelligence and useful connected capabilities into one workspace across Windows, macOS, Linux, Android, and iOS.
Whether you are studying on a laptop, developing software on Linux, researching on macOS, working from a Windows workstation, speaking to your phone, analyzing an image, reviewing a document, or running a local project agent, OpenMindAI is designed to give you more control over how AI becomes part of your work.
Local when possible. Connected when useful. Controlled by you.
Windows · macOS · Linux · Android · iOS
Website: https://www.openmindai.org
Start with the latest published release, then choose where your models and local data should live.