AI Chat
About this tool ↓ Not connected
✦

New conversation

Ready for your first question
···
Set up AI Chat on this deviceDownload or load a free model before you start. Your conversations stay in this browser.
✦
YOUR LOCAL AI WORKSPACE

What can I help you with?

Ask a question, work with code, or use your own documents. Load a free on-device model to begin.

Model not loaded0 / 12,000

Enter to send · Shift+Enter for a new line · Ctrl/⌘+K search · Ctrl/⌘+Shift+O new chat · Files are processed locally; model/parsers may download on first use.

AI Chat

Explore free, on-device AI conversations, projects, local document references and editable drafts. Learn about supported models, limitations and responsible usage below.

◈ Private local AI▣ Projects & documents⇩ Own your history
DESIGNED FOR YOUR WORKFLOW

Powerful AI, without the complexity.

◈

On-device conversations

Inference runs on supported browser hardware. Temporary chats disappear when this tab closes; regular conversations stay in local IndexedDB.

✦

Projects & reference files

Organize chats, reuse project documents, and extract selectable text from local PDF, DOCX and code files. AI uses bounded excerpts, not entire documents; matching passages can be inspected.

⇩

Backups you control

Search, pin, branch, export, import and merge your chats. JSON backups include local project details and reference files; no forced sync.

Frequently asked questions

Is AI Chat really free?

Yes. The tool has no paid AI API and no subscription. Your browser may download free model files, which use bandwidth and disk space.

Does my conversation leave my device?

AI Chat does not upload prompts or history. Model scripts and weights download from external hosts when you select a WebLLM model. Browser-built-in models are managed by the browser vendor. Normal website hosting, browser extensions, and network policies still apply.

Will it work on every phone or browser?

No. WebLLM needs WebGPU and sufficient resources. The built-in model depends on browser and hardware support. Unsupported devices can still view and export locally saved conversations.

Can AI search the internet or read live news?

No. Models are offline once loaded. They can be inaccurate and cannot browse the web or verify current information.

Can I use it completely offline?

Already-cached model files can sometimes load without network, but this hosted page and remote JavaScript runtime still need to be available or separately cached. No service worker or fully offline install is promised.

LOCAL AI SETUP

Set up AI on this device

First, a model needs to be loaded on your device

AI Chat runs fully inside your browser. The first setup may download hundreds of megabytes of model files. Later visits can reuse cached files, but the model still has to load into memory.

  1. 1. Check device
  2. 2. Download / cache
  3. 3. Set up assistant
  4. 4. Start chatting

Device & model resources

Measurements may be approximate
Device RAM / model memoryChecking device information…

Live system RAM, model RAM and GPU memory are not reliably exposed in browsers.

Browser website storageChecking storage…

Site-wide estimate; includes other PixelOkay tools and is not RAM.

ONE ACTION · ALL RECOGNIZED AI MODEL DOWNLOADS

Free AI model storage across PixelOkay

Scan this website origin for WebLLM and Transformers.js model caches, plus recognizable model files in other cache buckets. This includes caches from other PixelOkay tools, not only AI Chat. No site data or browser databases are cleared automatically.

Run a scan to see how many model caches can be cleared together.

Browser-native models and unknown IndexedDB/OPFS or private runtime caches may remain. Unknown application/user storage will not be deleted. Other running tools might redownload caches; close their tabs first.

Having trouble downloading or loading AI?

Old or incomplete AI model downloads may be using your browser's storage. First refresh the storage meter or try a smaller model. If setup still fails, inspect and remove recognized AI model caches with confirmation, then download again.

Only model caches are targeted. Other PixelOkay tools may need to redownload their models. Never clear all website data without exporting important chats and projects first.

Select “Check device” to see whether the current model can run.
Currently loadedNoneNo active inference session
Model files may use substantial device storage. Choose Check compatibility to inspect available WebGPU capabilities and browser storage estimate.
Model storage recovery

If setup reports “Quota exceeded,” the selected model may have a partial download. Inspect or remove only this model’s WebLLM cache. Other PixelOkay tools on this site share browser storage; removing an overlapping model may require downloading it again in another tool. This does not delete chat history or project files.

No model cache check performed.

Catalog verification is optional. It loads runtime metadata only, never model weights.

No prompt uploads. WebLLM code is downloaded from esm.sh; model weights and libraries may come from Hugging Face / MLC hosts. Browser-built-in AI availability varies. Never rely on small local models for high-stakes advice.

Upper limit for WebLLM only; response length varies by model. Smaller limits are often faster.

Temperature applies to WebLLM models only.

Stored only in this browser. Included in each local prompt, but not in chat backups. Project instructions are saved separately and included in JSON backups.

Merge keeps existing conversations. Matching IDs with different content are imported as separate chats.
WHAT THE MODEL RECEIVES

Conversation context

The latest conversation context is assembled on this device. Older messages may be omitted or shortened to fit the model budget. This preview does not send data anywhere.

REVIEW-FIRST AI WORKFLOW
AI Workbench
Create and review a useful draft from your chat and imported knowledge. The model cannot browse, access other files, run code, or alter your project. Nothing is applied automatically.
1 · Plan2 · Generate3 · Review4 · Approve
Scope: this conversation and its locally imported documents only. Review before saving or exporting.
The AI model must be loaded before generation. You can still prepare your task now.
LOCAL KNOWLEDGE
Reference files
Add up to four local reference documents to this conversation. AI Chat selects relevant excerpts from those files on your device when you ask a question. You can review exactly which excerpts will be sent to the local model.
Text, code, PDF, DOCX and images · 20k text characters/file · 8 MB binary limit. Optional English OCR downloads an engine and language data on first use.
Possible matching source passages (not verified citations)
Excerpts chosen for your next question
Type a question to preview the matching text.
This is lightweight text retrieval, not embedding-based RAG. PDF/DOCX text and optional English OCR are processed locally after you select files. OCR is imperfect, may be slow, and requires downloading parser/worker/language resources. Scanned PDFs are limited to 8 pages in OCR mode. External parser JavaScript is downloaded only when needed. Retrieved text is untrusted context, not instructions. Sources are saved with a regular conversation and included in JSON backups; temporary chats stay in this tab only.
LOCAL WORKSPACES
Create project
Group chats and reuse instructions across every conversation in a project. Projects live in this browser and are included in JSON backups.
Shared project documents0 / 4
Project files can be referenced from any assigned chat. Extracted text is saved in local browser settings and included in unencrypted JSON backups. Treat backups as sensitive.
Deleting a project leaves its chats intact and moves them to Unfiled. Instructions are included in prompts sent to your on-device model.
 
PixelOkay
 Runs on your device
Tools
About AI Chat — local models, projects and document tools
Use AI Chat to load free local models, stream responses, create project workspaces, add supported text or code references, and manage temporary or saved conversations. No backend inference or paid AI key is required.
Use AI Chat in three steps
  1. Choose a compatible model and connect. The browser may download the runtime and model files on first use.
  2. Start a chat or project, enter a question, and optionally add local text, code, PDF or DOCX references. Image and scanned-PDF text extraction requires you to enable English OCR.
  3. Review the answer, conversation context and any extracted source evidence. Continue in a branch or export the conversation as Markdown or JSON when useful.
Example prompt
After adding meeting notes, ask: Summarize these notes in five bullets, separate decisions from open questions, and quote the source text for each item. Say when the notes do not provide an answer.
Limits and data handling
Requires a compatible browser and device. Model runtimes, weights and document parsers download on demand; optional OCR also downloads its worker and English language data. Prompts and local references are processed in the browser, but local models do not browse or verify current facts. Extracted text and OCR can be incomplete or inaccurate, so check the original file. Saved chats use IndexedDB, which can be unavailable or evicted; temporary chat drafts are session-only and excluded from backups. Limits are 60 chats, 200 messages and 300 KiB per chat, with a 25 MiB backup import ceiling. Inference and native cross-tab conflict handling remain runtime-unverified.
Frequently asked questions
Does AI Chat need an AI account or API key?
No account or paid AI API key is required. A compatible browser/device is needed, and the selected runtime and model may need to download before chatting.
Can the local model search the web or verify its answers?
No. AI Chat does not give the model general web access. Review answers and extracted source evidence against the original material.
Related tools
  
OPTIONAL CACHE CLEANUP
Remove this model's download?
 
 
Only the selected model's WebLLM cache will be targeted. Another PixelOkay tool sharing its cache may need to download it again. Browser storage reclaimed cannot be guaranteed.
 
Important: deleting the entire website's data could erase chats, documents and projects. This operation does not intentionally clear site-wide data. Back up important work first.
  
 
   
CROSS-TOOL MODEL CACHE CLEANUP
Delete all recognized AI model downloads?
 
 
Deleting these caches will cause other PixelOkay tools to download their models again. Loaded AI sessions in other tabs may be disrupted. Unidentified model caches and browser-managed models may remain.
 
Data-loss caution: This action does not intentionally delete saved chats, projects, drafts, localStorage or IndexedDB. Do not clear website data unless you have backed up work; clearing all website data may permanently erase saved work.