Windows and Android previews available
Your AI assistants. Your workspace. Local first.
Build applications, prepare project costings and turn research into polished deliverables. LocalFirst-AI brings your chosen AI model, tools and project workspace together, so an assistant can carry out the work, check results and prepare the handover.
One assistant—or a coordinated team. Use hosted or local models. Give the project a clear objective, choose the tools it needs and follow progress from the first brief to the final result.
From first brief to final handover
Ask for a working application, a checked costing or a sourced launch package. With a suitable model and the right tools, an assistant can create files, test its work and refine the result while you stay focused on the outcome.
Software builder
“Build an internal service-desk app with staff login, tickets and a dashboard. Add tests, resolve build failures and include setup documentation.”
Start with a writable project folder, the required SDKs and build tools. Review the application before deployment.
Explore the application exampleEstimating assistant
“Use these scope documents, quantities, rates and prices to create an auditable workbook, risk scenarios and a client-facing PDF.”
Supply the scope, quantities, rates and tax assumptions. Review estimates and unresolved gaps before sharing a client quote.
Explore the costing exampleResearch and delivery
“Turn this product brief into a sourced market assessment, positioning, a presentation, a budget and a launch plan.”
Supply a product brief and permitted research sources. Approve purchases, external communications and publication separately.
Explore the launch exampleYour brief defines the scope. Review important results, set provider-side budgets and keep LocalFirst-AI running while the assistant works. Get started with independent project work.
Organised around your work
Create an assistant with the expertise and tools your work needs. Keep its role ready for future projects, with a dedicated space for each brief, conversation and set of deliverables.
Assistant
Choose a model and give your assistant a role, working style and goals. Add the tools it needs for coding, research or business projects.
Room
Keep your assistant's instructions, preferred model and tools together in a Room. Use that Room to organise related projects and return to the same assistant.
Workspace
Give each project its own conversation, notes, tasks and deliverables. Workspaces keep related work separate, with instructions, models and tools tailored to the task.
Model profile
Save a connection to a hosted service or local model server. Reuse it across projects and switch models without re-entering the connection details.
A place for every project
Keep a product assistant in one Room, then give the application, costing and launch plan their own workspaces. Each project has a clear brief and a place to keep its results.
Reusable Skills
Skills turn writing styles, research methods and release checklists into reusable guidance. Create your own or import a trusted package, then choose which instructions belong in each Room or workspace.
Create and keep
Keep your writing style, checklists and templates in a Skill. Or turn a useful response into a workflow draft you can refine and use again.
Choose and request
Let the assistant choose suitable guidance, ask for a particular method, or leave a Skill out of a project. A picker beside your message makes it easy to choose.
Inspect and share
Give each specialist the methods its assignment needs. Open Skills loaded on a response when you want to review the guidance supplied.
Use trusted Skills, review important results and remove confidential details before sharing. Your connected models and services may charge for usage. Read the Skills guide.
Explore what's new in 0.6.1 Preview 1.
Specialist teams · Introduced in 0.5.0 Preview 1
Give one assistant the objective and let it coordinate a team. Supervisor workspaces bring focused assignments, specialist results and shared project limits into one view.
The Supervisor coordinates the project, assigns focused work to specialist assistants and brings their results together. Each specialist works in its own workspace, so you can follow progress, review findings and access the deliverables.
Plan a product launch with a research lead and a presentation specialist. The research lead can coordinate separate competitor and customer reviews, while the Supervisor brings the findings and presentation together.
Example team · Product launch
Assigns the work, reviews findings and combines deliverables.
Brings the market research together.
Turns the agreed findings into an editable presentation.
Set how many specialists can work together and how much time and model usage their assignments allow. In 0.6.1 Preview 1, the Supervisor can finish reviewing results and preparing a handover after specialists reach their limits.
In 0.6.1 Preview 1, stop the assignment and its team straight from the message box. Keep active work separate from finished history, and turn off Supervisor to hide the panel without losing saved results.
Delegation preserves the project's tool and file-tool access rules, but workspaces do not isolate code or connected services. Use trusted tools and provider-side budgets; request limits do not cap charges.
Make it your assistant
Create a software builder, research assistant or estimating colleague with clear goals and instructions. Reuse that role across projects, or choose project-specific instructions when the work calls for a different approach.
Choose the tools each assistant and project may use. Keep everyday tools readily available, with other enabled capabilities easy to find through the searchable catalog.
Tools for the whole project
Manage files, build software, research a question and prepare documents with more than 200 built-in tool operations. You choose which tools each assistant may use; availability depends on your platform, access settings and connected services.
Files and software
Explore a codebase, edit files, refactor C# code and run builds and tests. Supported desktop tools can also work with Visual Studio.
Documents and data
Create Word documents, Excel workbooks, PowerPoint presentations and styled PDF reports on Windows. Keep editable source files alongside charts, comparisons and finished documents.
Research and context
Search the web, read source pages and compare findings. Return to previous conversations or use semantic memory to find relevant saved context. Web research requires a separate Brave Search API key.
Work management
Keep priorities, requirements and follow-up tasks alongside the project. You and the assistant can update notes and tasks as the work develops.
Computer and device
Where supported, let the assistant read screen content and interact with apps using the keyboard and mouse. You choose when desktop interaction is appropriate.
Extensible catalog
Connect trusted MCP servers to use external tools alongside LocalFirst-AI's built-in capabilities. Choose the tools each assistant needs for its project.
Explore extensionsYour model, your choice
Switch between hosted services and compatible local models while keeping your assistant's role and saved project work. Model profiles let you reuse each connection across Rooms and Workspaces.
Explore GPT-6 Astra, GPT-6.1 Sol, Claude Opus 5.5 or Sonnet 5.5, or a local coding model such as KAT-Coder-V2.5-Dev, subject to provider access and server support. Choose a model that handles your project's tools and requirements reliably.
Check model access and supported settings with your provider, then try a small tool-enabled task before starting a larger project. The examples above are not performance guarantees.
Choose your setupBeyond text
Add image, speech and memory capabilities as your projects need them. Each can use a specialist model alongside the one handling your main conversation.
Create artwork, refine an existing image and use the results in documents or presentations with a compatible image model.
Listen to responses with a voice, speed and style you choose, using a compatible speech profile.
Dictate a message or use realtime audio for a spoken conversation, where your provider supports it.
Find useful saved context by meaning, even when you do not remember the exact wording, with an embeddings profile.
Connect your existing systems
Use Model Context Protocol (MCP) servers to connect your assistant to external tools and services. Choose a trusted hosted server or run a compatible server on your desktop.
Select which tools each Room or Workspace may use, so every project has the connections it needs.
Connect to remote tools using the server's supported authentication.
Use trusted tool servers running on your desktop.
Choose the external tools needed by each assistant and project.
Your model. More than one kind of work.
Keep software, research and business projects in one working environment. Choose your models, reuse your assistants and coordinate specialist teams while keeping the project's decisions and deliverables together.
Work with OpenAI, Claude, Gemini or compatible local models. Choose the connection that suits each assistant or project.
Save the role, instructions and tools for a software builder, researcher or estimating assistant, then reuse them on your next project.
Review sources, decisions, tasks and deliverables alongside the conversation. Update notes and priorities yourself as the project develops.
Bring implementation, research, workbooks and reports into the same project, with focused tools and access for each workstream.
Codex and Claude Code also support autonomous coding. LocalFirst-AI offers a workspace that extends across software, research and business deliverables, with your choice of models, reusable assistant roles and shared project records.
Explore the independent-assistant guideStart with one model
Follow guided setup to connect a model and assign it to Chat. Utility is optional: Chat handles supporting tasks unless you configure a separate profile. Create your first Room, choose the project's tools and send a clear brief.
OpenAI API account
https://api.openai.com/v1gpt-6-astra or gpt-6.1-solUse the Responses API for tool calling with these models. Connect with an OpenAI API project key and confirm that your account has model access.
OpenAI setup guideAnthropic API account
https://api.anthropic.comclaude-opus-5-5 or claude-sonnet-5-5Connect with your Anthropic API key and choose Opus or Sonnet. The Claude setup guide explains which settings to use with your model.
Claude setup guideLocal or self-hosted
http://localhost:1234/v1Choose a tool-capable model supported by your server. KAT-Coder-V2.5-Dev is text-only, with 35B total and 3B active parameters; check its hardware requirements before downloading.
Local model setup guideMore connection options
Connect a Google AI API key through the Gemini profile, or use your Azure resource endpoint and deployment name with the model's supported API mode.
For other compatible services, use the connection details and model ID supplied by your provider.
Explore provider setup guidesEarly access
The current release is 0.6.1 Preview 1. Windows x64 offers the widest range of local file, build and document tools. Android is available as an early sideloaded preview for version 7.0 and later. Web, macOS and iOS packages are not available in this release.
Connect a compatible hosted service or a running local model server. LocalFirst-AI does not include model downloads, API keys or credits. Start with a small task and choose the tools and access that suit your work.
Find official downloads, release notes and setup guides in the Preview repository, along with privacy, security and support information.
Visit the official preview repositoryWindows and Android previews
Get the Windows installer or portable archive for desktop. For Android 7.0 and later, download the APK or the bundle with supporting preview documents.
Before you begin
Learn how to choose a model, organise projects and stay in control of your assistant's tools, data and costs.
Read the full preview scopeYes, with a capable model, a clear brief and the right tools and access. An assistant can create files, implement features, calculate costs and check its work. You still need to review important results and approve consequential actions. Applications need testing before production use, and estimates need review before becoming client quotes.
Yes. Skills in 0.6.1 Preview 1 keep your methods ready for the next task. Create your own, import trusted guidance or turn a useful response into a workflow draft. Let the assistant choose a suitable Skill or ask for a particular one from the message box. Read the Skills guide.
It can. Choosing suitable Skills and working with their guidance can contribute to normal model usage. Charges depend on your connected providers, even when the main Chat model is local. Set provider-side budgets and monitor usage. Read the billing guide.
Yes. Enable Supervisor delegation in Workspace settings to coordinate specialist assistants in the same Room. Allow nested Supervisors when a workstream needs a team of its own. Follow progress in Supervisor assignments and open each workspace to review its conversation and deliverables. Read the delegation guide.
Set shared limits for workers, nested teams, concurrent work, model requests and duration. Request limits do not cap charges, so also use provider-side budgets. Workers cannot broaden inherited tool or file-tool access, but workspaces do not sandbox shell commands or external services. Use trusted tools, assign separate file responsibilities and review the results.
In 0.6.1 Preview 1, Stop in the message box ends this workspace's assignment and its team. Resume continues work already paused for input; it does not undo a cancellation. Wait for work to close before reusing the workspace. Pausing does not extend deadlines, and stopping does not undo completed changes or charges. Interrupted work does not restart automatically. Check the guide for your installed preview.
Yes, an active text response can continue while you visit another workspace. Use the assistant activity menu to return to it or stop it. LocalFirst-AI must remain running: closing the app can interrupt work. Review saved progress and any completed external actions before starting another attempt.
No. Connect your own API account or local model server. Providers may charge separately for model requests, search, media or infrastructure, and consumer chat subscriptions may not include API access. Longer projects and teams can make many requests; set provider-side budgets and monitor usage. Read the billing guide.
No. LocalFirst-AI connects to the service or local server running your model. It provides the assistant roles, tools and project workspaces around that connection. Install or host a local model separately, or use a supported provider's API.
The model generates responses and tool requests. Your assistant adds a role, instructions and selected tools to guide that work. Change the model when needed while keeping the assistant's role and saved projects.
A Room is the home for an assistant and its default instructions, model and tools. Workspaces organise individual projects within the Room, each with its own conversation, notes, tasks and deliverables. You can tailor instructions, model choice and tools to each workspace.
Use a short role description or a detailed system prompt. Workspaces use the Room's instructions unless you give them their own, so you can keep a consistent assistant or tailor its approach to a particular project.
Enable or disable tool groups and individual operations for each Room or Workspace. Selected tools can be Direct, while other enabled tools remain searchable through the Gateway. Either route uses the same availability rules; it does not grant additional file access. Some tools also require a supported platform or connected service.
Your assistant's role, tools and saved work remain in place. Choose another compatible profile as a default or for a particular Room or Workspace. A provider change starts a fresh model session; it does not delete your saved conversations, notes or deliverables.
Add a compatible profile for each capability you want: images, speech, realtime audio or embeddings for semantic memory. These can use different models from your main Chat profile, so you can add specialist capabilities as needed.
Yes. Connect a compatible MCP server over HTTP/SSE, or a trusted local server over stdio on desktop. Then choose which of its tools each Room or Workspace may use. Review the server's access and data policies before connecting it.
LocalFirst-AI keeps Rooms, Workspaces, notes, tasks and local application data on your device. Content included in model requests or sent to connected services follows those services' data policies. A local Chat model does not make the whole project offline if another capability or tool uses a remote service.
Windows x64 and Android are available as public previews. Android requires version 7.0 or later, with arm64-v8a device and x86_64 emulator support. Windows offers the widest local file, build and document tool coverage. Web, macOS and iOS packages are not available in this release; connecting Android to a remote model does not add Windows desktop tools.
Download the official APK, verify its SHA-256 checksum and open it from your browser or file manager. Android may ask you to allow installation from that app; older versions use a device-wide setting. Restore the setting after installation. The preview bundle includes the same APK and supporting guides.