Skip to content
Locale
Get Started
BROWSER. DEVICE. CONNECTED MODELS.

The local
AI harness.

Built for small local workloads and large, long-term projects. Learn agentic coding for free in Chrome, or use the Agent for multi-model orchestration.

Get Started Download Agent

View pricing

Use local models within Locale to help the environment. Smaller models can use less energy and run without cloud compute.

Build the homepage Scratch History
1 FILE
index.htmlUpdated
Create a homepage with a project gallery.

I’ll create the page in your workspace and open a preview.

Used 1 toolWrote index.html

The homepage is ready to review.

Used 1 toolPreviewed index.html
Preview opened. Files saved in your workspace.
Message Locale…
Agent Qwen3 8B · Ollama
Illustrative workflow · results vary by model
MODELS AND CONNECTIONS
Chrome AI Ollama OpenCode Grok Kickbacks.ai
THE HARNESS

Small tasks.
Long-term projects.

A harness connects models to context, tools, and a workspace. For large-scale projects, use the Agent to coordinate model groups, delegate tasks, and retain project context.

Multi-model workflows

Choose a model per chat, consult another model, or delegate a bounded task. Keep each model’s output and context visible.

Reusable Agent Groups

Save a primary model, advisors, and shared instructions. Apply the group to project chats for a consistent workflow.

Projects and workspaces

Organize chats, files, and tasks. Build in a browser Sandbox with live previews, or work in a Device folder with the Agent.

Plan, Build, and Ducky

Plan an approach, implement with tools, or discuss a problem in read-only Ducky mode. Fork chats to explore alternatives.

Controlled tool access

Review command approvals and scope file access to selected folders. Inspect tool activity, changed files, and milestones.

Kickbacks.ai support

Earn money from optional sponsored placements through Kickbacks.ai. Use local models to reduce paid inference costs. Earnings vary.

Search and persistent chat history Context and usage tracking Browser files, previews, and ZIP export Deep completion and task tracking

Model consultation, Agent Groups, connected providers, Kickbacks.ai, and Device tools require Locale Agent. Browser features remain available in the free trial.

Workspace and output features

Import text and binary files, edit code, preview HTML and app output, inspect the browser console, and export source as a ZIP. Browser snapshot history supports commits and branches. Device workspaces use actual disk folders and real Git through the Agent.

Conversation and task features

Search chats and projects, pin and archive conversations, fork a discussion, inspect reply versions, queue messages, and track tasks and milestones. Saved notes and context compaction help retain relevant information. Deep completion and bounded loops support longer work.

Model and image features

Choose and favorite models, compare observed throughput, track context usage, and set supported reasoning controls. Image attachments and image descriptions depend on model capabilities. With the Agent, consult other models or delegate bounded tasks with separate transcripts and inherited permissions.

Device and compute features

The Agent adds file and command tools, provider credentials in the operating system credential manager, model management, GPU/VRAM readings, and encrypted local-network pairing with reconnect support. Experimental opt-in workers handle independent text requests on trusted local networks; they do not combine GPU memory into a single model.

LOCALE AGENT

Your browser.
Your machine’s capabilities.

Keep working in Locale from the comfort of your browser. The Windows Agent connects local Ollama models, project folders, approved commands, model groups, and Kickbacks.ai.

Pair your phone or another browser on your trusted local network to work with the same machine. The Agent runs the tools and model connections while your browser stays the interface.

Download for WindowsRelease notes

Windows x64 · Portable · Keep the Agent open and your machine online. Current phone pairing uses your trusted local network; internet device access is planned.

  1. 1

    Download and open

    Run LocaleAgent.exe. The Agent stays available in the system tray.

  2. 2

    Open your workspace

    Open Locale in your browser. Pair another browser or your phone with the Agent’s code on your trusted network.

  3. 3

    Extend your workspace

    Connect local or provider models, select a project folder, and use tools from the browser. Review command approvals and keep the Agent open.

LOCAL FIRST

Start in Chrome.
Extend with the Agent.

A free starting point for beginners learning agentic coding, with orchestration tools for power users managing larger projects.

01 / BROWSER

Run entirely in Chrome

Learn agentic coding with Chrome’s on-device Gemini Nano foundation model. Create files and preview small projects for free, with no Agent, account, or API key.

  • Gemini Nano inference on your device
  • History and Sandbox files in your browser
  • No connected model providers
Get Started Supported desktop Chrome required. Initial model download may be needed.
02 / DEVICE

Work on your machine

Use model groups and task delegation for larger projects. Locale Agent adds real project folders, approved commands, local Ollama models, and optional provider connections.

  • Local Ollama models and device tools
  • Model groups and provider connections
  • Paired access on your trusted local network
Download Agent Agent features require the Agent to remain open.
Choose local compute when it fits.

Chrome and local Ollama models run on your hardware. Smaller models can reduce resource use and reliance on remote servers. Environmental impact depends on hardware, workload, and energy source; carbon savings have not been measured.

MODEL CHOICE

Use the right model
for each part of the work.

Start with a local model. Add a specialist for review or a group for repeatable tasks. You control which providers receive your requests.

Chrome foundation model

Gemini Nano, directly in supported desktop Chrome.

Local · Free

Ollama

Run installed local models through the Agent on your own hardware.

Local · Agent

OpenCode

Connect supported Zen or Go models with your own credentials.

Connected · Agent

Grok

Choose an xAI API key or supported Grok Build subscription sign-in.

Connected · Agent

Examples of supported models

Use installed local models or connect a provider through the Agent. These examples are separate from Cloud’s upcoming model catalog.

Ollama

Llama 3.2Qwen3 8BGemma 3DeepSeek R1

Local models from Locale’s model library. Hardware requirements vary.

Provider details ↗

OpenCode Zen / Go

Kimi K2.6MiniMax M2.7GLM-5.1

Examples from connected provider catalogs. Access depends on your account.

Provider details ↗

Provider availability, account requirements, usage limits, and fees apply separately. The free Agent supports your own provider connections. The approved plans add account-based inference credits to Sync and Cloud. Keep uses Chrome’s foundation model or your Agent connections. Cloud account inference is still in development.

AGENT GROUP EXAMPLE

A local primary. A second opinion when needed.

Save model roles and instructions once, then reuse them across project chats. Consultation requests go only to the models you select.

OllamaPrimary+OpenCodeAdvisorYour projectShared context
TOOLS AND FUNCTIONS

Work with files, code,
and project context.

Locale exposes tools according to the model, workspace, and work mode. Plan and Ducky restrict changes. Build carries out work with the required permissions.

Browser + Agent

Read, search, and edit files

List a workspace, find matching text, read selected lines, and write changes. Browser files stay in Sandbox; Device files stay within the folders you select.

Functions
list_dirread_fileread_linesload_filegrep_filewrite_file
Browser workspace

Run and inspect previews

Run JavaScript, open a page preview, inspect its controls, and test interactions. Supported image models can describe a preview so the model can review the result.

Functions
scratchrun_javascriptpreview_pageinspect_previewwebsite_preview
Browser workspace

Save versions and export

Review changes, save snapshots, create branches, and export your files. Browser snapshot history stays separate from real Git repositories on your machine.

Functions
sandbox_gitbrowser_export
Browser + Agent

Track tasks and project context

Maintain a task list, record milestones and project notes, check context usage, and ask for decisions when work needs your input.

Functions
task_listtask_updatemilestoneproject_notecontext_usageask_user
Multiple models via Agent

Consult and delegate

Request a second opinion, delegate a bounded task, and collect its result. Reusable Agent Groups define the primary model, advisors, and shared instructions.

Functions
consult_modeldelegate_taskagent_tasks
Agent required

Run device commands

Run shell, PowerShell, or Bun commands in your project and initialize real Git repositories. Commands follow Locale’s approval rules and workspace permissions.

Functions
run_commandrun_powershellrun_buninit_git
Agent + Ollama

Manage local models

Check installed models, download an Ollama model, and send work to a local model. Model setup and inference run on the computer hosting your Agent.

Functions
ollama_statusollama_pullollama_chat
Depends on model and tools

Use images and web information

Describe image attachments with a supported model. The Agent adds web search when current information is needed. Web access requires an internet connection.

Functions
describe_imageweb_search

The free Chrome version uses the foundation model and browser tools. Agent connections add local and connected models, device tools, and Kickbacks.ai. Sync and Cloud include account inference credits at launch. Cloud adds a worker. Keep uses Chrome Nano or your own model connection. Cloud services are in development.

PLANS

Store, sync, and build.

Local use is free. Choose the cloud services you need.

Cloud services in development. Mock checkout only, no charges.

Keep

Storage + site

Store your work and publish a personal site.

$5/ month

$5 every month

10 GB cloud storage

Site files and databases use this storage.

  • Personal site
  • 100 MB site database*
  • 100 monthly active site users*
  • 25 monthly site credits*
  • 12,000 weekly site asset requests
  • Browser workspaces and optional Agent

Use Chrome Nano or your own model connection.

  • No device sync
  • No Box VM worker
  • No included inference credits
Choose Keep

Sync

Models + sync

Account inference, websites, and device sync.

$10/ month

Monthly subscription

50 GB cloud storage

Site files and databases use this storage.

  • 3 shareable sites
  • Device sync
  • 300 monthly inference credits ($ 3)
  • 250 MB site database*
  • 500 monthly active site users*
  • 50 monthly site credits*
  • 35,000 weekly site asset requests
  • Browser workspaces and optional Agent

Use cloud models through your Locale account. Device tools require an open Agent.

  • No Box VM worker
Choose Sync

Cloud

Models + worker

Models and a worker for independent cloud tasks.

$25/ month

Monthly subscription

256 GB cloud storage

Site files and databases use this storage.

  • 10 shareable sites
  • Device sync
  • 800 monthly inference credits ($ 8)
  • 60 small Box VM hours / month
  • 1000 MB site database*
  • 2,000 monthly active site users*
  • 200 monthly site credits*
  • 115,000 weekly site asset requests
  • Browser workspaces and optional Agent

One worker: 2 shared vCPUs, 4 GB RAM, at least 12 GB working space. Pauses when idle.

    Choose Cloud

    Inference through your account. Sync and Cloud include model access. 100 inference credits cover $1 of published model usage. Tokens used for input, output, and reasoning draw from the same balance. Model availability varies by provider. Device files and local models require an open Agent.

    Sites with room to grow. Publish at a shareable HTTPS URL. *Database, site sign-in, and site credit allowances are planned launch targets. They cover managed backend functions, not arbitrary server processes. Access is static-only.

    100 site credits represent $1 of metered backend usage, separate from inference. Planned optional top-up: $5 for 400 site credits. No automatic charges. Site request limits reset each Monday at 00:00 UTC. Credits, active-user counts, and worker hours reset monthly without rollover. Purchased site credits are planned to last 12 months.

    Hosting and usage limits

    Allowances are per account across its sites. Each site supports up to 100 MB, files up to 20 MB, counted within storage. Site files, database contents and indexes count toward your cloud storage allocation. Database scans, writes, and function execution consume site credits. New work pauses at the limit. Site accounts, metering, publishing, and cloud services are not active yet.

    Use Locale locally for free

    The browser harness and Windows Agent remain free.

    Get Started
    COMMON QUESTIONS

    Before you start.

    What can I do without the Agent?

    The static browser version uses only Chrome’s foundation model. It supports chat and browser workspaces, including files, previews, tasks, and export. It cannot connect to Ollama, external model providers, device folders, Kickbacks.ai, or remote sync.

    Does the browser version work on every device?

    No. It needs desktop Chrome with an available Prompt API and supported hardware. Chrome may download its model the first time you use it. Unsupported browsers show setup information instead of sending your prompt to a cloud fallback.

    Where does my data go?

    Browser-only chats and workspace files are saved in your browser. Local Ollama inference runs on your machine through the Agent. If you select an online provider, that provider receives the context sent to it. Cloud storage will be a separate subscription service.

    Does local mean I can work offline?

    Downloaded Chrome and local Ollama models can perform inference offline. The website and required assets must already be available. Downloads, connected providers, web tools, Kickbacks.ai, and future cloud services need an internet connection.

    What does Cloud add?

    Cloud is $25/month for 256 GB, sync, ten static sites, 800 monthly inference credits ($8 of model usage), and 60 small Box VM hours. Cloud services remain in development. Agent Groups and your own provider connections remain available with the free Agent.

    How is Sync different from Cloud?

    Sync is $10/month with 50 GB, device sync, three static sites, and 300 monthly inference credits ($3) through your Locale account. Cloud adds a larger model budget and a Box worker. Device tools still need the Agent. These offers are not active yet.

    How does Keep pricing work?

    Keep is $5 every month for 10 GB and a Personal site, without device sync or inference credits. Model tasks require Chrome’s foundation model or an Agent connection.

    What happens when I close the Agent?

    Agent tools and local model connections stop being available. Keep the Agent open and the host online for its models, device tools, and paired access. The browser version runs independently. Planned Cloud work can continue without a local Agent.

    Can I subscribe today?

    You can create an account now. Approved offers are available to review in mock checkout. No payment is collected or cloud service activated. The browser version and Agent remain free.

    Runtime documentationChrome Prompt API Ollama local inference