RangerIO Guide

9. Frequently Asked Questions

General Questions

What is RangerIO?

RangerIO is a desktop application that brings AI-powered data analysis capabilities to your local machine. Unlike cloud-based solutions, all data processing and AI inference happens entirely on your computer - nothing is sent to external servers.

Is RangerIO really 100% offline?

Yes. After the initial installation and model downloads:
- ✅ No internet connection required
- ✅ No raw data leaves your machine without your explicit action
- ✅ No telemetry or analytics
- ✅ No license validation calls
- ✅ No automatic updates

Network access happens only when you opt in:
- Download additional AI models
- Check for software updates manually
- Route a query through the Stage 2 connector to a cloud AI tool (sensitive values are masked before they leave; restored when the answer returns)
- Import data from external APIs or databases (your choice)

Is RangerIO free? What's the licensing?

The Lite edition is free for individual professionals — full data engine, dossier building, Privacy Scan, and Markdown export. You bring your own AI (local GGUF on disk, or a cloud subscription via the Stage 2 connector).

The Full edition bundles a local LLM runtime so RangerIO can answer locally without any external model.

Enterprise licensing is available for teams that need additional capabilities. The bundled local models ship under their respective open-weight licenses (Apache 2.0, MIT, Llama Community License) — all commercially permissible. Cloud models you connect via Stage 2 are governed by your subscription with that provider.

How does RangerIO compare to cloud AI tools?

RangerIO doesn't replace cloud AI tools — it makes them safe to use on more of your work. ChatGPT, Claude, and Perplexity are excellent at general reasoning but have no awareness of your data and no way to handle sensitive content. RangerIO profiles your data locally and bridges it to the AI tools you already use, sending the dossier — not the raw rows — and masking sensitive values along the way.

Feature RangerIO Cloud AI alone
Data Privacy Raw data stays local; only the masked dossier travels Whatever you paste is sent to the provider
Compliance posture Easier to fit into HIPAA / GDPR / SOX workflows Requires provider certifications + careful policy
Internet Required Optional — Stage 2 connector is opt-in Yes, always
Performance Depends on your hardware (local) or network + provider (cloud) Depends on network + provider
Cost Free Lite edition; Full and Enterprise are paid Usually subscription per provider
Data Size Limits Your hardware limits Service limits and context windows
Customization Full control over models, PII rules, and routing Limited to provider settings

Installation & Setup

What are the minimum system requirements?

Lite (BYO model):
- 16GB RAM
- Modern CPU (no GPU required)
- ~5GB for the base install plus 2-8GB per local model
- macOS 12+ (Apple Silicon or Intel) or Windows 10/11

Full (bundled local LLM):
- 32GB RAM on x86, or Apple Silicon with 16GB+ unified memory
- Discrete GPU with 8GB+ VRAM (NVIDIA CUDA, or AMD Vulkan on Windows) — optional but faster
- macOS 12+ or Windows 10/11

Do I need a GPU?

No. The Lite edition works on CPU only. The Full edition is faster with a GPU but doesn't require one. Apple Silicon uses Metal automatically; NVIDIA uses CUDA; AMD on Windows uses Vulkan. CPU fallback is always available.

Which GPU should I get?

For local inference, more VRAM is the main lever. Useful breakpoints:

Entry (small models, 1-4B):
- NVIDIA RTX 3060 12GB
- Apple Silicon M-series with 16GB+ unified memory

Recommended (medium models, 7-8B):
- NVIDIA RTX 4070 12GB or RTX 4080 16GB
- Apple M-series Pro/Max with 16GB+ unified memory

Heavy use (14B+ models, frequent inference):
- NVIDIA RTX 4090 24GB
- Apple Silicon Ultra with 32GB+ unified memory

How much disk space do I need?

Breakdown:
- Base application: ~5GB
- AI models: 1-15GB each (you choose which to download)
- Your data: Depends on your datasets

Planning guide:
- Minimal setup: 15GB (app + 1-2 small models)
- Typical setup: 30GB (app + 3-4 models + workspace)
- Comprehensive setup: 100GB+ (app + many models + large datasets)

Can I install on an external drive?

Yes! During installation, you can choose where to store:
- Application files (usually on internal drive)
- AI models (can be on external drive)
- Project data (can be on external drive)

Note: External drives may be slower, especially USB 2.0 drives. Use USB 3.0+ or Thunderbolt for best performance.

How do I update RangerIO?

RangerIO does not auto-update (by design, for privacy).

To update manually:
1. Check current version: Help → About
2. Visit RangerIO website to check for new version
3. Download new installer
4. Run installer (your data and settings are preserved)

Data & Privacy

Does my data ever leave my computer?

Raw data, dossiers, Privacy Scan results, and PII mappings stay on your laptop by default. Data leaves only when you take an explicit action:
- Routing a query through the Stage 2 connector — sensitive values are masked before any cloud round-trip, and the mapping back to real values stays on your device
- Exporting data to a network location
- Connecting to an external database or API (your choice)

RangerIO never sends data anywhere without your explicit action.

Is there any telemetry or tracking?

No. RangerIO collects zero telemetry. No usage statistics, no crash reports (unless you manually generate and send them), no analytics.

Can I use RangerIO in a compliance-regulated environment?

RangerIO's architecture is designed for regulated workflows — local-first processing, governed cloud handoff, opt-in connectivity, no telemetry. But "HIPAA-ready" or "SOC 2-certified" are certifications that belong to your organization and your audit posture, not to a piece of software. Use RangerIO as part of your compliance program; it does not substitute for one.

Common frameworks it can support:

HIPAA (Healthcare):
- Raw PHI stays local; only masked dossiers ever touch the network
- Privacy Scan with confidence tiers
- Override decisions persist across rescans

SOX (Financial):
- Local export audit trail (what was sent, what was masked, when)
- No external data transmission unless you opt in

GDPR / CCPA (Privacy):
- Data minimization through Privacy Scan and masking
- You control deletion — data lives in your project directory
- No third-party data processors by default

Export-controlled / government:
- Air-gapped deployment option (disable Stage 2 connector)
- No cloud dependencies

Note: Always consult your compliance officer to ensure RangerIO fits your specific requirements.

How is sensitive data protected?

At rest:
- Pair RangerIO with OS-level full-disk encryption (FileVault on macOS, BitLocker on Windows)
- Secure file permissions on the project directory

In use:
- Data engine and Privacy Scan run on your laptop
- When you use the Stage 2 connector, sensitive values are masked before any cloud round-trip and restored when the answer returns
- The mapping between masked stand-ins and real values lives only on your device
- Every export records what was sent, what was masked, and when — a local audit trail you control

AI & Models

Which AI models does RangerIO support?

Any GGUF-compatible llama.cpp model. Families that run well today:
- Qwen3 (Alibaba) — 0.6B to 14B parameters
- Qwen3.5 (Alibaba) — 0.8B to 27B+ parameters
- Phi-3.5-mini (Microsoft) — ~3.8B parameters
- DeepSeek-R1 Distill — 1.5B, 7B, 14B parameters

Cloud models are reachable through the Stage 2 connector (Claude Desktop today; ChatGPT and Perplexity in named phases).

What's the difference between the models?

Model size affects:
- Quality: Larger models = better understanding and reasoning
- Speed: Smaller models = faster responses
- Memory: Larger models = more RAM required

General guidance:
- 1-3B models: Fast, good for simple queries, low memory
- 3-7B models: Balanced performance, recommended for most users
- 7B+ models: Best quality, slower, need 16GB+ RAM and preferably a GPU

What is quantization?

Quantization reduces model size by using fewer bits per number:

Q4 (4-bit):
- Smallest: ~25% of original size
- Fastest inference
- Slight quality loss (usually acceptable)
- Best for: Limited RAM, CPU-only systems

Q5 (5-bit):
- Medium: ~31% of original size
- Good performance
- Minimal quality loss
- Best for: General use

Q8 (8-bit):
- Larger: ~50% of original size
- Slower but higher quality
- Almost no quality loss
- Best for: When you have RAM to spare and want best quality

Can I use my own / custom models?

Currently, RangerIO supports models in GGUF format. If you have a custom model:
1. Convert to GGUF format (using llama.cpp tools)
2. Place in RangerIO models directory
3. RangerIO will auto-detect it

Support for additional formats may be added in future versions.

Why is the AI slow on my machine?

Common causes:

1. CPU-only inference
- Solution: Add a GPU, switch to a smaller model, or route the query through the Stage 2 connector

2. Large model on limited hardware
- Solution: Use a smaller model or higher quantization (Q4 instead of Q8)

3. Insufficient RAM
- Solution: Close other applications or upgrade RAM

4. Old hardware
- Solution: Check minimum requirements, consider upgrade

5. Deep Search mode
- Solution: Use Basic or Assistant mode for faster responses

The AI gave an incorrect answer. What should I do?

AI models can sometimes make mistakes. Here's what to do:

1. Verify the data
- Check if the information exists in your data
- View the source chunks the AI referenced

2. Try a different approach
- Rephrase your question
- Use Deep Search mode for better accuracy
- Try a larger model

3. Report (optional)
- Note the question and incorrect answer
- This helps improve future models

4. Cross-check important answers
- For critical decisions, always verify AI responses against source data

Data Import & Export

What file formats can I import?

Structured data:
- CSV, TSV (any delimiter)
- Excel (.xlsx, .xls)
- JSON, JSONL
- Parquet

Databases:
- PostgreSQL
- MySQL / MariaDB
- Microsoft SQL Server
- SQLite

Documents:
- PDF (text extraction)
- Microsoft Word (.docx)
- Plain text

APIs:
- REST endpoints
- JSON data sources

What's the largest file I can import?

Practical limits:
- Files: Limited by available RAM (typically can handle files up to 1/4 of your RAM)
- Databases: No limit (query-based)
- PDF/Documents: Up to 10,000 pages recommended

Examples:
- 16GB RAM → ~4GB file max
- 32GB RAM → ~8GB file max
- 64GB RAM → ~16GB file max

For larger datasets, use database connections or split files.

Can I import data from cloud services (Dropbox, Google Drive, etc.)?

Yes, but indirectly:
1. Download files from cloud service to local machine
2. Import files into RangerIO

RangerIO itself doesn't connect directly to cloud services (by design, for privacy).

Can I export to cloud services?

Yes, similar to import:
1. Export data to local file
2. Upload file to cloud service

Or:
- Export to database (can be cloud-hosted database)
- Mount cloud drive locally, export there

How do I import large datasets efficiently?

Best practices:

For very large CSV files:
- Import in chunks (split file)
- Or use "streaming import" option

For databases:
- Use "live query" mode instead of importing
- Or sample data for exploration, full import for final analysis

For many small files:
- Use directory watching
- Or batch import (select all at once)

Performance & Optimization

How can I make RangerIO faster?

Short term (no hardware changes):
1. Use a smaller GGUF model (e.g., Qwen3 1.7B or Phi-3.5-mini)
2. Use higher quantization (Q4 vs Q8)
3. Use Basic RAG mode for simple queries
4. Close other applications (16GB Lite hardware can thrash if many apps are open)
5. Restart RangerIO occasionally to clear cached model singletons

Medium term (software/settings):
1. Update graphics drivers (for GPU acceleration)
2. Enable hardware acceleration in settings
3. Increase allocated RAM in settings
4. Use SSD instead of HDD for data storage

Long term (hardware upgrades):
1. Add or upgrade GPU
2. Add more RAM (16GB → 32GB)
3. Upgrade CPU
4. Use NVMe SSD

My GPU isn't being detected. What should I do?

Troubleshooting steps:

1. Check if GPU is supported
- NVIDIA: CUDA-capable (recent GeForce / RTX)
- AMD on Windows: Vulkan-capable
- Apple: M-series chips (M1+)

2. Update drivers
- NVIDIA: Download from nvidia.com
- AMD: Download from amd.com
- Apple: Update macOS

3. Confirm acceleration toolchain
- NVIDIA: A recent CUDA-capable driver
- AMD on Windows: A recent Vulkan-capable driver

4. Restart after updates

5. Check settings
- RangerIO → Settings → Performance
- Verify GPU is listed
- Enable "Use GPU acceleration"

6. Still not working?
- Check Help → System Info to see detected hardware
- Contact support with system info

How much RAM does each model need?

Approximate RAM requirements:

Model Q4 Q5 Q8
Qwen3 0.6B 2-3GB 2-3GB 3-4GB
Qwen3 1.7B 3-4GB 3-5GB 4-6GB
Qwen3 4B 5-7GB 6-8GB 7-10GB
Qwen3 8B 9-11GB 10-13GB 13-16GB
Phi-3.5-mini 4-6GB 5-7GB 7-9GB
DeepSeek-R1 Distill 7B 8-10GB 10-12GB 13-15GB

Note: These are minimum requirements. Add 2-4GB for the application and data.

Troubleshooting

RangerIO won't start

Common solutions:

1. Check system requirements
- 16GB RAM minimum (Lite); 32GB or Apple Silicon 16GB+ unified (Full)
- ~5GB for the install plus 2-8GB per local model
- macOS 12+ or Windows 10/11

2. Check for conflicting applications
- The backend runs on port 9000; the dev frontend runs on 5173
- Close other applications using those ports

3. Check logs
- Help → Open Logs Folder
- Look for error messages

4. Reinstall
- Uninstall RangerIO
- Delete config folder (backup first if needed)
- Reinstall fresh

Import fails with "Out of memory"

Solutions:

1. Close other applications
- Free up RAM

2. Import smaller chunks
- Split large files
- Use sampling

3. Increase virtual memory (swap)
- OS settings

4. Upgrade RAM
- If frequently encountering this issue

AI responses are slow

See "How can I make RangerIO faster?" above.

Export fails

Check:
- Enough disk space at destination?
- Write permissions to destination folder?
- Filename valid (no special characters)?
- Database connection working (if exporting to DB)?

Data quality checks keep failing

This isn't necessarily a problem! Quality checks are designed to find issues. If checks "fail":
1. Review specific issues found
2. Use Prepare Wizard to fix issues
3. Re-check quality

Only the most perfect datasets get A grades. B or C grades are common and usually acceptable.

Getting Help

Where can I find more documentation?

How do I report a bug?

Collect this information before reporting:

Then send the report through your support channel. Enterprise customers have a dedicated contact; everyone else can reach us via the website's support form.

How do I request a feature?

Include:
- What you want to do
- Why it would be useful
- How you envision it working

Send feature requests through your support channel or the website contact form.