RFlect AI Features — Status & Roadmap¶

Last Updated: May 12, 2026 Current Version: v4.2.0 Status: Beta / Enabled in GUI and MCP

Overview¶

RFlect ships AI-powered features for intelligent antenna analysis. A unified provider abstraction (plot_antenna/llm_provider.py) supports OpenAI, Anthropic (Claude), and Ollama (local models), so users can choose between cloud and on-prem inference without changing the rest of the toolchain.

There are two integration surfaces:

In-GUI — AI Chat Assistant + AI-powered report generation, accessible from the desktop app.
MCP server — rflect-mcp/server.py exposes 34 tools an AI agent (Claude Code, Cline, etc.) can call programmatically.

What Works¶

1. AI Chat Assistant¶

Status: ~85% Complete

Real-time conversational analysis of loaded measurements
Function calling: get_gain_statistics, analyze_pattern, compare_polarizations, generate_2d_plot, generate_3d_plot
Rich context awareness (loaded files, frequencies, data shape, key metrics)
Quick-action buttons in the GUI (Gain Stats, Pattern, Polarization, All Freqs)
Multi-turn history, multi-provider support

2. AI-Powered Report Generation¶

Status: ~90% Complete

Executive summary generation from loaded measurements
Automatic gain statistics insertion
Pattern type classification
Vision-capable plot analysis (provider-dependent)
Uses the same unified provider abstraction as the chat assistant

Pending: - Custom branding integration (partially implemented) - Multi-frequency comparison tables - Automated figure captioning - Compliance checklist generation (FCC, CE)

3. Multi-Provider Support (since v4.0.0)¶

Status: Complete

Provider	Tool Calling	Vision	Notes
OpenAI	GPT-4 (Chat Completions) + GPT-5 (Responses API)	GPT-4o+	Default
Anthropic	Claude Messages API	All Claude models	Via Tools → Manage API Keys
Ollama	llama3.1+, qwen2.5+	llava, llama3.2-vision	Local, no API key needed

4. Secure API Key Management¶

Status: Complete

Fernet AES-128 encryption with PBKDF2 key derivation (600K iterations)
Machine-ID based encryption key (/etc/machine-id, IOPlatformUUID, MachineGuid)
OS keyring integration (Credential Manager, Keychain)
Restrictive file permissions (chmod 600 / Windows ACL)
Keys stored in _key_cache dict instead of os.environ

5. MCP Server (34 tools)¶

Status: Complete

Category	Count	Examples
Import	6	`import_antenna_file`, `import_antenna_folder`, `import_passive_pair`
Analysis	5	`analyze_pattern`, `get_gain_statistics`, `compare_polarizations`
Reports	3	`generate_report`, `preview_report`, `get_report_options`
Bulk	5	`bulk_process_passive`, `bulk_process_active`, `list_measurement_files`
UWB	3	`calculate_sff_from_files`, `analyze_uwb_channel`, `get_impedance_bandwidth`
Calibration Drift	8	`cal_drift_ingest`, `cal_drift_compare`, `cal_drift_report`, `cal_drift_set_setup_group`
Orchestration	1	`process_folder` (new in v4.2.0)
Misc	3	`get_measurement_details`, `validate_file_pair`, `convert_to_cst`

`process_folder` (v4.2.0)¶

A single entry point that scans a folder, picks the right workflow (passive HPOL/VPOL pair, active TRP, cal-drift archive, or UWB sweep), runs it, and optionally produces a DOCX report. Replaces the previous list → bulk → analyze → report chain that MCP clients had to script manually.

process_folder(folder_path, intent='auto'|'passive'|'active'|'cal_drift'|'uwb',
               report=False, freqs=None, report_path=None)

Auto-detect priority: cal_drift > passive > active > uwb. Mixed folders proceed with the winner and surface a mixed_intents_detected warning. The tool never raises — every failure mode (missing folder, no match, partial pair, per-file UWB error, report write failure) returns as a structured warning.

Known Limitations¶

Pattern Analysis — HPBW and F/B ratio implemented and verified (boundary wrapping + IEEE-validated TRP). No sidelobe-level detection or symmetry analysis yet. Pattern classification limited to 3 categories.
Function Calling Compatibility — works across all major model families. Older Ollama models may have limited tool support.
Ollama Vision — only vision-capable models (llava, llama3.2-vision, gemma3) support plot analysis; others fall back to text-only.

Architecture¶

plot_antenna/
├── llm_provider.py         # Unified provider abstraction
├── ai_analysis.py          # Pure analysis logic (GUI-independent)
│   └── AntennaAnalyzer     # Reusable analysis class
├── gui/
│   └── ai_chat_mixin.py    # AI chat GUI integration
├── save.py                 # Report generation
└── api_keys.py             # Secure key management

rflect-mcp/
├── server.py
└── tools/
    ├── import_tools.py
    ├── analysis_tools.py
    ├── report_tools.py
    ├── bulk_tools.py
    ├── uwb_tools.py
    ├── cal_drift_tools.py
    └── orchestration.py    # process_folder (v4.2.0)

Configuration¶

Enabling AI¶

Pick one provider: 1. OpenAI — API key via Tools → Manage API Keys 2. Anthropic — ANTHROPIC_API_KEY env var 3. Ollama — running locally, no key required

Select provider via Tools → AI Settings.

Disabling AI¶

If no provider is configured: - AI Chat menu item is hidden - Generate Report with AI falls back to template-only mode - Core RFlect functionality continues to work normally

Models Supported¶

OpenAI: GPT-4o-mini (default), GPT-4o, GPT-5-nano, GPT-5-mini, GPT-5.2
Anthropic: Claude Sonnet, Claude Opus
Ollama: llama3.1, qwen2.5, mistral, llava (vision), and more

Roadmap¶

Shipped — v4.0.0 (Feb 2026)¶

Architecture refactor (mixin-based GUI)
Multi-provider AI (OpenAI, Anthropic, Ollama)
Secure API-key management (Fernet, OS keyring, machine-ID binding)
AntennaAnalyzer with HPBW, F/B ratio, batch analysis
11 RF engineering formula fixes
MCP server with 20 tools

Shipped — v4.1.x (Feb–Apr 2026)¶

Maritime / horizon antenna plots (5 plot types)
Advanced RF analysis suite (Link Budget, ITU-R, Multipath, MIMO, Wearable/SAR)
Calibration drift tracker with cross-epoch comparison
Active Chamber Calibration regression fix

Shipped — v4.2.0 (May 2026)¶

process_folder MCP orchestrator (single-call folder workflows)
UWB analysis helper extracted from MCP wrapper for reuse
MCP server expanded to 34 tools

Planned — v4.3+¶

AI datasheet extraction (vision-based parameter extraction from PDF/images)
Sidelobe detection and reporting
Automated figure insertion in reports
Complete branding integration
Multi-frequency comparison tables
Simulation vs measurement comparison
AI-powered anomaly detection
Integration with electromagnetic simulation tools

For Developers¶

Testing AI Features¶

from plot_antenna.ai_analysis import AntennaAnalyzer

data = {
    'phi': phi_angles,
    'theta': theta_angles,
    'total_gain': gain_array,
    'h_gain': h_pol_array,
    'v_gain': v_pol_array,
}

analyzer = AntennaAnalyzer(
    measurement_data=data,
    scan_type='passive',
    frequencies=[2400, 2450, 2500],
)

print(analyzer.get_gain_statistics(frequency=2400))
print(analyzer.analyze_pattern(frequency=2400))
print(analyzer.compare_polarizations(frequency=2400))

Adding New Analysis Functions¶

Add function to AntennaAnalyzer in ai_analysis.py
Expose to the chat assistant via gui/ai_chat_mixin.py
Add tests under tests/test_ai_analysis.py
Wrap as an MCP tool under rflect-mcp/tools/analysis_tools.py
Update this document

Contributing¶

See CONTRIBUTING.md.

Support¶

Issues: https://github.com/RFingAdam/RFlect/issues
Discussions: https://github.com/RFingAdam/RFlect/discussions

Tag AI-feature issues with ai.

Disclaimer¶

AI-generated analysis and recommendations are provided as guidance only. Always verify results with established RF engineering practices and measurements. RFlect's AI features are experimental and should not be relied upon for critical design decisions without human expert review.

RFlect AI Features — Status & Roadmap¶

Overview¶

What Works¶

1. AI Chat Assistant¶

2. AI-Powered Report Generation¶

3. Multi-Provider Support (since v4.0.0)¶

4. Secure API Key Management¶

5. MCP Server (34 tools)¶

process_folder (v4.2.0)¶

Known Limitations¶

Architecture¶

Configuration¶

Enabling AI¶

Disabling AI¶

Models Supported¶

Roadmap¶

Shipped — v4.0.0 (Feb 2026)¶

Shipped — v4.1.x (Feb–Apr 2026)¶

Shipped — v4.2.0 (May 2026)¶

Planned — v4.3+¶

For Developers¶

Testing AI Features¶

Adding New Analysis Functions¶

Contributing¶

Support¶

Disclaimer¶

`process_folder` (v4.2.0)¶