User Guide — What Can Frank Actually Do For You?

Type anything. "What's the weather in Vienna?" Frank searches the web and tells you. "What's 15% of 89.99?" He runs Python and gives you 13.50 — not a guess, an actual calculation. "What's on my screen?" He takes a screenshot and describes what he sees.

Frank isn't a search engine. He's an agent. He doesn't just answer — he does things. Every response might use tools: web search, file operations, calculations, screenshots. You'll see tool indicators in the chat when he's working.

Type / in the chat. A menu appears. 39 commands, from /weather Vienna to /screenshot to /compose (write an email with AI). Slash commands are shortcuts — everything they do, you can also ask for in natural language. But they're faster.

Key ones to try first:

• /weather — current weather for any city
• /screenshot — Frank sees your screen and describes it
• /timer 25 — Pomodoro timer
• /note Buy milk tomorrow — saves to persistent notes
• /search quantum computing 2026 — web search
• /system — shows CPU, RAM, GPU, disk status

"Remember that my cat is called Mochi and she's 3 years old."

Done. Frank stores this in his long-term memory (Titan Memory System). Next week, next month — ask "what's my cat's name?" and he knows. Unlike every other AI, Frank doesn't forget when you close the tab. His memory lives in databases on your machine, independent of the language model.

You can also:

• "What do you remember about me?" — see everything stored
• "Forget that I told you about the dentist" — delete specific memories
• "Search your memory for cooking recipes" — targeted recall

Type: "Research quantum error correction"

Watch what happens:

1. Frank breaks your question into 3-5 sub-questions
2. He searches the web (DuckDuckGo) — 3 results per query, rate-limited
3. He searches ArXiv for academic papers
4. He fetches and summarizes the best web pages (RAG pipeline)
5. He checks his own Titan memory for relevant prior knowledge
6. If the topic matches a science domain, he runs lab experiments
7. He extracts key findings into a structured table
8. He writes an expert synthesis

The result: a styled PDF with table of contents, numbered sections, citations, and source links. It appears in ~/Documents/Frank/research/. Open the Laboratory page in the Web UI to see research history and download past reports.

This is not summarization. This is autonomous research with real sources.

Frank has a full simulation lab. Ask him to experiment:

• "Experiment: what angle maximizes projectile range?" → Physics station runs trajectory simulations
• "Experiment: what happens to a glider in Game of Life after 100 steps?" → GoL sandbox
• "Experiment: calculate eigenvalues of this matrix..." → Math station with sympy
• "Experiment: RC circuit with 10kΩ and 100μF, show charging curve" → Electronics station
• "Experiment: simulate quantum tunneling through a 2eV barrier" → Quantum lab with actual state vector math
• "Experiment: 3-body gravitational system with Earth, Moon, and Sun" → Astronomy orrery with leapfrog integration
• "Experiment: train a GAN on digit patterns" → Adversarial neural network training

Each station runs real math — numpy, sympy, PyTorch. Not language model guesses. Budget: 20 experiments/day.

The Hypothesis Engine uses these stations automatically. Frank generates hypotheses from idle thoughts, designs experiments to test them, interprets results, and revises his beliefs. This happens autonomously while you're away.

• "Write me a PDF with a weekly meal plan" → create_pdf with markdown formatting
• "Create an Excel budget spreadsheet with categories: Rent 1200, Food 350, Transport 80" → create_excel from CSV data
• "Write a Word document summarizing yesterday's meeting" → create_word with .docx output
• "Build me a Python script that renames all JPGs by date" → build_script — creates executable code

All files appear in ~/Documents/Frank/. Frank names them sensibly and tells you where they are.

Frank has universal file conversion powered by ffmpeg, ImageMagick, Pandoc, and LibreOffice:

Images: PNG ↔ JPG ↔ WebP ↔ BMP ↔ TIFF ↔ GIF ↔ ICO ↔ SVG

Audio: MP3 ↔ WAV ↔ OGG ↔ FLAC ↔ AAC ↔ M4A ↔ WMA

Video: MP4 ↔ AVI ↔ MKV ↔ WebM ↔ MOV ↔ WMV ↔ GIF

Documents: DOCX ↔ PDF ↔ HTML ↔ Markdown ↔ EPUB ↔ LaTeX ↔ TXT ↔ RTF

Cross-format: Video → Audio (extract soundtrack), PDF → Images, Image → PDF

Just say: "Convert ~/Downloads/report.docx to PDF" or "Extract audio from ~/Videos/lecture.mp4 as MP3"

• "Summarize ~/Documents/contract.pdf" → file_summarize reads and condenses
• "Deep analysis of ~/Documents/thesis.txt" → doc_ingest breaks into hierarchical chunks, summarizes each, produces a structured overview
• "Map the code structure of ~/project/main.py" → code_map lists classes, functions, imports

"Transcribe ~/Downloads/meeting.mp3" → Whisper speech-to-text (INT8, runs on CPU)

Supports: MP3, WAV, OGG, FLAC, M4A, AAC, WMA, MP4, MKV, WebM, AVI, MOV, WMV. Auto-detects language.

Open the Mail page. Configure your email provider (Gmail, Outlook, or custom IMAP/SMTP) via /mailconfig.

Frank reads your inbox, shows unread messages, and lets you:

• Read any email — full content rendered in the UI
• Reply with AI — Frank reads the conversation, drafts a reply in your chosen tone + language
• Compose — describe what you want to say, Frank writes it
• Search — find old emails by keyword

8 languages: English, German, French, Spanish, Italian, Portuguese, Japanese, Chinese 4 tones: Professional, Casual, Friendly, Formal

Frank never sends without your approval. Every draft is editable.

Frank can bridge to messaging platforms:

Telegram: Set up a bot token → Frank receives and responds to Telegram messages through his regular chat pipeline. Configure via the Web UI bridges panel.

WhatsApp: Similar bridge — start/stop via API. Frank processes WhatsApp messages the same way he processes chat input.

These are experimental features. They work, but they're not as polished as the Web UI chat.

Hotkey: Ctrl+Shift+F — opens a crosshair overlay. Select a region. Frank analyzes it.

Chat: Drag & drop any image into the chat window. Or click the camera icon.

Two modes:

• Describe — Full pipeline: CNN scene classification → DINOv2 understanding → OCR text extraction → LLM interpretation. ~100ms.
• Transcribe — OCR only. Extracts text from screenshots or photos. English + German.

Frank sees: text, objects, faces, scenes, UI elements. He can describe what's on your screen, read error messages, or identify what app you're using.

Push-to-talk: Hold Space (when chat input isn't focused) to record. Release to send.

Frank listens (faster-whisper, INT8), thinks, and speaks back (Piper TTS for German, Kokoro TTS for English). ~155 MB RAM, CPU-only.

Safety: buffers shorter than 0.3 seconds are ignored (prevents accidental taps). Empty transcripts show a friendly "hold longer and speak clearly" toast.

• "How much disk space do I have?" → runs df -h
• "What processes are using the most CPU?" → runs top equivalent
• "Open Firefox" → app_open|firefox
• "Close Firefox" → app_close|firefox
• /usb → list connected USB devices
• /network → network interfaces, IP addresses
• /system → full system health dashboard

All shell commands run in a Firejail sandbox. A blocklist prevents dangerous operations (no rm -rf, no dd, no network attacks).

• "Remind me to call the dentist tomorrow at 3 PM" → persistent reminder, survives restarts
• "Set a 25-minute pomodoro timer" → countdown with notification
• "Add meeting with Sarah on Friday at 14:00 for 1 hour" → calendar event
• /calendar → today's schedule
• /week → this week's schedule
• /todos → all active reminders and tasks

"Save a note: project deadline is April 15" → saved to persistent note system

"Search my notes for deadline" → finds it

Notes are separate from memories. Notes are things you explicitly save. Memories are things Frank infers and stores from conversation context. Both persist forever.

The main page. Chat input at bottom, messages above. The right sidebar scrolls Frank's consciousness stream — his idle thoughts, mood shifts, and activity. You can literally watch him think in real time.

Supports: text, images (drag & drop or camera button), voice (Space to talk), slash commands, and all 45 tools via natural language.

40 glowing spheres in WebGL 3D space. Each sphere = a running Linux service. Lines = data connections. Colors = mood. Speed = cognitive load. Crashed services go dark.

Hover any sphere → port number, uptime, memory usage, CPU%, last activity. Click → details panel.

Based on Reynolds flocking (1987) with mood noise — when Frank is stressed, the swarm scatters.

Masonry layout of Frank's generated artwork. HUD corners, scan-sweep hover effect. Each piece has creation metadata.

Research trigger: type a topic, watch 8 phases light up, get a PDF. Shows experiment history, hypothesis engine status, active hypotheses, and 7 simulation stations.

IMAP inbox, AI-drafted replies, composition with tone/language control. See the email section above for details.

The wiki you're reading right now (also on DaemonWeave). 49 articles, full-text search, cross-linking.