First Launch Checklist
Before connecting any accounts, confirm the basics are working.
- Open the app and confirm it launches without errors.
- Check that the version matches what you intended to install.
- Verify the Memory Tree vault path is set to a location with sufficient disk space.
- Confirm your preferred theme (light/dark) and language settings.
- Test the search function to ensure the local index is accessible.
Connecting Your First Integration
Start with one low-risk connector to understand the OAuth flow.
- Go to Settings > Integrations and pick a service you use daily (e.g., Gmail or Google Calendar).
- Click Connect and review the OAuth permission screen carefully.
- Accept only the minimum scopes needed for your use case.
- Wait one sync cycle (~20 minutes) and check the Memory Tree tab for ingested data.
- Open the Obsidian vault to inspect what was stored.
Configuring Model Routing
Model routing determines which AI provider handles each task. Understanding this helps you control costs and privacy.
- Cloud providers (default): heavy reasoning, coding, vision, and voice tasks route to frontier models via the bundled subscription.
- Local AI (optional): embeddings, summary-tree building, and lightweight tasks can run on Ollama or LM Studio.
- Path: Settings > AI & Skills > Model Routing.
- You can override the default provider and base URL in config.toml.
- Vision tasks always route to a vision-capable model regardless of local/cloud settings.
Setting Up Local AI
Local AI is optional but recommended for privacy-sensitive workflows.
- Install Ollama or LM Studio separately.
- In OpenHuman: Settings > AI & Skills > Local AI.
- Choose a preset: embeddings only, memory plus reflection, or full local where supported.
- The default local model is Gemma 3 1B, requiring a few gigabytes of RAM and disk.
- Verify local AI is working by checking that memory embedding tasks appear in the Ollama logs.
Memory Tree Configuration
Fine-tune how the Memory Tree behaves.
- Vault path: choose a location that is backed up and has sufficient space.
- Sync frequency: default is ~20 minutes. Reduce for battery life; increase for near-real-time updates.
- Chunk size: default is 3,000 tokens. Reduce for more granular retrieval; increase for context efficiency.
- Cleanup: periodically review the vault for outdated or irrelevant chunks and delete them.
Voice and Mascot Setup
Enable the desktop mascot and voice features for a more interactive experience.
- Grant microphone permission in system settings.
- Enable voice in OpenHuman: Settings > Voice & Mascot.
- Whisper STT runs locally for speech recognition.
- ElevenLabs TTS requires an API key and routes to the cloud.
- The mascot can join Google Meet as a guest participant with real-time transcription.
Verification Steps
Before relying on OpenHuman for daily use, run through these checks.
- Ask a question about recent emails and verify the answer references actual content.
- Check that deleted emails are removed from the Memory Tree after a sync cycle.
- Test a local AI query and confirm it does not hit the cloud subscription.
- Verify that revoked connector tokens are handled gracefully (no crashes, no stale data).
- Confirm the mascot responds to voice commands if enabled.
Troubleshooting First Setup
If something goes wrong during initial setup, these are the most common issues.
- OAuth hangs: check network, proxy, and system clock accuracy.
- Memory tab is empty: confirm the vault path is writable and the first sync completed.
- Local AI not detected: verify Ollama is running on port 11434.
- Voice not working: check microphone permissions and ElevenLabs API key.
- Missing permissions on macOS: grant Accessibility, Screen Recording, and Microphone in System Settings.