Skip to main content
While jo runs locally on your Mac, you can optionally call in cloud models when you need extra computing power or specialized capabilities.

Available cloud models

jo supports the most powerful AI models available:

GPT (OpenAI)

GPT-4o and GPT-5.2 for advanced reasoning and analysis

Claude (Anthropic)

Claude 3.5 Sonnet for complex tasks and long-form writing

Gemini (Google)

Gemini Pro for multimodal understanding and analysis

Grok (xAI)

Grok for real-time information and reasoning
Local model included: jo also ships with a local model that runs entirely on your Mac. Swap between models anytime.

When to use cloud models

Most queries run completely locally on your Mac. Use cloud models when you need:

Complex reasoning

  • Multi-step problem solving
  • Strategic planning
  • Analyzing trade-offs
  • Complex decision-making
Example: “Help me prep for this salary negotiation” — Claude can roleplay scenarios and suggest strategies.

Advanced analysis

  • Deep document analysis
  • Code generation and review
  • Technical explanations
  • Research synthesis
Example: “Review this contract and explain the key terms” — GPT can provide detailed legal analysis.

Specialized tasks

  • Writing and editing
  • Mathematical calculations
  • Language translation
  • Creative brainstorming
Example: “Help me write a compelling product launch email” — Claude excels at long-form writing.

Extended context

  • Processing many documents at once
  • Long conversation history
  • Connecting distant pieces of information
Local first: Try local queries first. They’re free, instant, and completely private. Use cloud models when you need that extra power.

When local is enough

Your Mac’s local model handles:

Search and retrieval

  • Finding emails, notes, or files
  • Searching your photo library
  • Looking up past conversations
  • Locating documents
Example: “Where’s that W-9 from my accountant?” — Local model finds it instantly.

Summaries

  • Summarizing email threads
  • Catching up on groupchats
  • Meeting note summaries
  • Quick document overviews
Example: “Catch me up on the family chat” — Local model summarizes locally.

Simple questions

  • Quick factual lookups from your data
  • Calendar queries
  • Contact information
  • Recent activity
Example: “When’s my next meeting?” — Local model answers instantly.
Privacy: Local queries never leave your Mac. Your personal data stays completely private when using the local model.

How to use cloud models

Call in a cloud model by choosing it before or during your conversation.

Selecting a model

  1. In the query box: Click the model selector
  2. Choose your model: Pick GPT, Claude, Gemini, or Grok
  3. Ask your question: jo sends that conversation to the cloud

What gets sent

When you use a cloud model:
  • Only that conversation is sent
  • Context you provide (any files or data you reference)
  • Not your full index — your entire data stays local
You control context: jo asks what information to include before sending to a cloud model. You explicitly choose what context gets shared.

Example workflow

  1. Local search: “Find all my notes about the Q4 strategy”
  2. jo finds them using the local model
  3. Switch to Claude: “Now help me write a presentation deck”
  4. You confirm: jo asks “Send these 5 notes to Claude?”
  5. Claude analyzes: Only those notes are sent, not your entire data

Pricing model

Cloud models use pay-per-use pricing. No subscriptions.

How it works

  • Local queries: Always free
  • Cloud queries: $0.01-0.05 per conversation
  • No monthly fee: Pay only for what you use
  • No commitment: Stop anytime

Average user

$5-10/month for occasional cloud model use

Power user

Under $20/month even with frequent cloud queries

Cost examples

Query typeModelTypical cost
Simple searchLocalFree
Groupchat summaryLocalFree
Document analysisGPT-4o~$0.02
Long-form writingClaude~$0.03
Code generationGPT-4o~$0.02
Complex reasoningGPT-5.2~$0.05
Most users spend $3-8/month: Because most queries run locally for free, you only pay for the advanced tasks that truly need cloud power.

Smart routing

jo automatically routes between fast and reasoning models based on your question.

Fast models

For most queries:
  • GPT-4o (OpenAI)
  • Claude 3.5 Sonnet (Anthropic)
  • Response time: 1-3 seconds

Reasoning models

For complex problems:
  • GPT-5.2 (OpenAI)
  • Claude 3.5 Opus (Anthropic)
  • Response time: 10-30 seconds
  • Higher accuracy on hard problems
Automatic selection: jo picks the right model speed based on your query complexity. You can also manually override if needed.

Connecting cloud models

Set up cloud models in jo’s preferences.

First-time setup

  1. Open jo preferences
  2. Go to Models tab
  3. Add API key: For OpenAI, Anthropic, etc.
  4. Test connection: jo verifies it works
  5. Start using: Cloud models now available

API keys

You’ll need API keys from:
Your own keys: jo uses your personal API keys. You pay the providers directly at their standard rates. No markup from jo.

Billing

You’re billed directly by the model provider:
  • OpenAI bills your OpenAI account
  • Anthropic bills your Anthropic account
  • Google bills your Google Cloud account
jo never handles payment. You pay the providers at their standard rates.

Privacy with cloud models

When you use a cloud model, that conversation is sent to the provider.

What’s sent

  • Your question
  • Context you explicitly chose to include
  • Previous messages in that conversation

What’s NOT sent

  • Your full data index
  • Other conversations
  • Your entire photo library or email archive
  • Anything you didn’t explicitly include
Local data stays local: Even when using cloud models, your full indexed data never leaves your Mac. Only the specific conversation you’re having is sent.

Provider privacy policies

Cloud providers have policies about data usage:
  • OpenAI: Doesn’t train on API data (per their terms)
  • Anthropic: No training on customer data
  • Google: Doesn’t train on Gemini API data
No training: Cloud inference uses providers with contracts that prevent training on your data.

Best practices

1. Start local

Try the local model first. It’s free, instant, and private.
❌ Don't: Immediately use GPT for "Find that email from my lawyer"
✅ Do: Use local model to find it, then GPT if you need analysis

2. Be explicit about context

When using cloud models, clearly state what context to include.
❌ Don't: "Analyze this" (unclear what "this" is)
✅ Do: "Analyze the Q4 budget spreadsheet and last month's email thread"

3. Use the right model

Pick the model that fits the task:
  • Simple search: Local model
  • Quick analysis: GPT-4o
  • Long-form writing: Claude
  • Code generation: GPT-4o
  • Complex reasoning: GPT-5.2
Let jo choose: If unsure, let jo auto-route to the right model. It’s optimized for cost and performance.

4. Review context before sending

When jo asks to send context to a cloud model, review what’s included:
  • Are you sharing sensitive information?
  • Is all this context necessary?
  • Can you be more selective?

5. Use local for sensitive data

For highly sensitive data (legal, medical, financial):
✅ Use local: Search and summary of sensitive documents
❌ Use cloud: Deep analysis of sensitive documents
Or use cloud models only for non-sensitive parts of your workflow.

Switching models mid-conversation

You can switch models during a conversation:
  1. Start local: “Find my notes on the product roadmap”
  2. Switch to Claude: “Now help me write a memo about this”
  3. Back to local: “Find the email thread with stakeholders”
  4. Switch to GPT: “Draft responses to their concerns”
Seamless switching: jo maintains context when switching between models. Your conversation flows naturally.

Offline usage

Cloud models require internet. The local model works offline.

When offline

  • ✅ Search your data
  • ✅ Summarize groupchats
  • ✅ Find files and emails
  • ❌ Cannot use GPT, Claude, Gemini, or Grok
Travel mode: The local model works anywhere, even on a plane. All your data is already indexed locally.

Future models

jo will support new models as they’re released:
  • New versions from OpenAI, Anthropic, Google
  • Additional model providers
  • Specialized models for specific tasks
You don’t need to update jo for new models — they’ll appear automatically when available.
Stay updated: jo checks for new available models and adds them to your model selector automatically.

Build docs developers (and LLMs) love