User Guide

Intellistream Machine Intuition is a real-time conversation intelligence tool that analyzes live discussions, predicts conversational trajectories, and provides strategic insights as you speak. It builds a mental model of your counterparts to anticipate their next moves.

Program Architecture

Intellistream is built as a multi-layered system designed for ultra-low latency conversational intelligence. The workflow moves from local audio capture through a centralized intelligence proxy to persistent storage and knowledge injection.

Frontend (Java Swing)

The desktop application provides the user interface for monitoring sessions, managing contacts, and viewing real-time insights. It is managed by the MainFrame and AppController classes.

Input Layer

  • Microphone Capture: Local audio processing via MicrophoneCaptureThread.
  • Meeting Bot: A Recall.ai-powered bot that can join remote meetings (Zoom, Google Meet, Teams) to capture audio.

Intelligence Layer (Intellistream Proxy)

A centralized backend that handles heavy lift operations:

  • Diarized Transcription: Identifying who is speaking in real-time.
  • LLM Inference: Using high-performance models like Gemini or GPT-OSS to analyze context.
  • State Management (ConversationState): Keeps track of the current transcript, participants, predicted branches, and strategic summaries.

Persistence Layer (StorageService)

  • Sessions: Saved in .intellistream/sessions/ as JSON snapshots for later review.
  • Participant Souls: Persistent Markdown files (soul.md) in .intellistream/profiles/ that capture long-term participant characteristics and behavioral patterns.

Knowledge Base (DocumentService)

Injects external context (PDFs, DOCX, TXT) into the AI's prompt at three levels: Global, Participant, and Session. This ensures the AI is grounded in your specific data.

Interface & Layout

The main window is organized into four primary sections to provide focused, real-time intelligence without overwhelming the user.

1. Global Controls (Top Header)

Located at the very top, this area manages the application's core state and background knowledge.

  • AI Pulse (Light Indicator): Located next to "Pulse Tree".
    • ● Amber: Processing data.
    • ● Green: Successfully matched live conversation to a predicted branch.
    • ● Gray: Standby / Ready for next turn.
  • Pulse Tree (Button): Manually forces the AI to analyze the current transcript and generate new predictions. Use this if the conversation moves faster than the 10s auto-pulse.
  • Start Meeting (Button): The "Master Switch." Opens selection for using your local microphone or deploying a remote meeting bot.
  • Library & Settings: Manage background knowledge (PDFs/Docs) and application configuration respectively.

2. Strategic View (Upper Section)

This area is for high-level visualization and long-term planning.

  • Inference Tree (Upper-Left): The large central panel taking up the top-left quadrant. A graphical tree that "grows" as you talk. Branches are "what-if" scenarios the AI predicts. If you say something the AI expected, the tree shifts and highlights that path in green.
  • Participants (Far-Left Column): The vertical list on the leftmost edge. Lists identified people and syncs with their "Soul" profiles to provide personalized context.
  • Strategic Sidebar (Right Column):
    • Strategic Summary: Summarizes the overall "vibe" and status of the negotiation/discussion.
    • Forecasted Path: Shows a sequence of cards representing the most likely future turns, including the specific Intent and Details.
    • AI Monitor: Takeaways: A running log of critical points that have been definitively settled or mentioned, acting as an automated executive summary.

3. Tactical View (Lower Section)

This area provides real-time data and immediate suggestions.

  • Live Transcript (Lower-Left Tab): The main text area at the bottom left. Displays diarized text (who said what). It serves as the raw ground truth for all AI logic.
  • Performance Review (Lower-Left Tab — Secondary): Appears after a session ends. It evaluates your strategy, highlighting tactical wins and missed opportunities in a comprehensive report.
  • Strategic Insights (Lower-Right Column):
    • Tactical Suggestions: Specific actions (e.g., "Ask about budget", "Validate their concern").
    • Potential Pivots: Emergency or strategic exits if the current topic is stalling.
    • Mindset Notes: The "Emotional Radar." It alerts you to underlying feelings or biases the AI detects (e.g., "Stakeholder seems skeptical of the timeline").
    • Active Alerts: Flashing/Red alerts for critical conversational failures (e.g., "Hostility detected", "Losing leverage").

4. Status Bar (Bottom Footer)

The thin strip at the very bottom. Shows the app's current activity (e.g., "Thinking...") on the left, and a Live Audio Level Meter on the right to confirm your voice is being captured correctly.

1. Download & Install

To begin using Intellistream, you must first download the platform client for your operating system. The client is distributed as a ZIP archive containing the executable binary and required assets.

  1. Head to the Download Section on the landing page.
  2. Sign in with your account to unlock the download buttons.
  3. Download the .zip file for your platform (Windows only; macOS/Linux not supported yet due to instability issues).
  4. Extract the contents of the ZIP file to a preferred directory on your machine.
  5. Run the Intellistream executable to launch the application.

Platform Specific Notes

Windows

Windows SmartScreen may flag the app as untrusted.

  • Click More info
  • Select Run anyway

macOS

Gatekeeper will block the app as "unnotarized".

  • Right-click the .app
  • Select Open
  • Click Open in the dialog

Linux

You must grant execution permissions to the binary.

  • chmod +x Intellistream
  • ./Intellistream

Once installed, proceed to the API & Proxy Setup section to configure your credentials.

2. API Key & Proxy Setup

  1. Visit the Dashboard to manage your account and API keys.
  2. Copy your API Key for use in the application settings. This key is required for all AI inference tasks.

3. First Launch & Performance

On the very first launch, the application will ask you to prioritize your performance tier:

  • Quality (Gemini): Uses gemini-3-flash-preview for the deepest strategic analysis and nuanced emotional tracking.
  • Speed (GPT-OSS): Uses openai/gpt-oss-20b for near-instant trajectory predictions and rapid-fire inference.

This choice sets your initial default model, which can be changed later in Settings.

4. Basic Configuration

Access configuration via the ⚙️ icon in the header or File → Settings:

  • API Key: Your unique key from the proxy dashboard.
  • Who are you?: In the desktop app, select your name in Settings > Proxy & API from the dropdown. This list is automatically populated from your Contact Directory and helps the AI provide more personalized strategic insights.
  • Model: Toggle between Gemini and GPT-OSS performance tiers.
  • Mixers: Select your primary Microphone and system audio if capturing remote participants locally. Not needed for remote meetings.

Starting a Meeting

Click the 🎙 Start Meeting button to begin capturing audio. You can choose between two modes:

  • Remote Meeting (Bot URL): [Primary Option] Enter a Google Meet, Zoom, or Teams link. A bot will join to provide high-fidelity transcription.
  • In-Person (Microphone): Captures audio directly via your local system hardware.

The Library & Document Upload

Click the 📖 Library button in the header. This is the central hub for providing context to the AI across three scopes:

  • Global System: Documents that provide context for every meeting (e.g., company standards).
  • Participants: Attach resumes or specific profiles to individuals (linked by name).
  • Current Session: Upload project docs or agendas specific to the active conversation.

Knowledge Flow:

  1. Open Library: Access the dialog via the header button.
  2. Select Scope: Choose the appropriate tab (Global, Participants, or Session).
  3. Add File: Upload your document (PDF, Docx, TXT).
  4. Human Explanation: (Crucial) Provide a brief note on how the AI should use this document.

The AI immediately incorporates this new context into the next "Inference Pulse."

Strategic Sidebar

  • Strategic Summary: Real-time analysis of power dynamics and conversational tensions.
  • Forecasted Path: AI's prediction of where the conversation is logically heading.
  • Takeaways: Actionable tasks or insights extracted automatically.

Tip: Panels will display "Waiting..." until enough conversational data has been gathered for synthesis.

Contact Directory

Access via Edit → Contact Directory. Here you can manage profiles for regular contacts as well as yourself. Important: After adding yourself to the directory, return to Settings > Proxy & API and select your name under "Who are you?" to help the AI provide better, personalized responses and view "Soul" documents—LLM-synthesized personality profiles that evolve with every session.

Configuration

All settings are stored in ~/.intellistream/config.properties:

Property Default Description
INTELLISTREAM_API_KEY (empty) Your API key from intellistream-proxy-371775672019.us-central1.run.app
PROXY_BASE_URL https://intellistream-proxy-... Proxy server URL
LLM Model openai/gpt-oss-20b AI model for inference
Audio Input Mixer (auto-detect) Microphone device name for in-person meetings
System Prompt / Context (default prompt) Custom system prompt for the AI
Max Prediction Width 3 Maximum branches per tree level
Max Prediction Depth 2 Maximum depth of the inference tree

Troubleshooting

Issue Solution
Missing Icons Ensure the application is run from a directory containing the required assets.
Bot Fails to Join Verify the meeting URL is correct and the meeting has started.
No Inference Check the AI Pulse light in the header. If it's grey, ensure your API key is valid and audio is flowing.
Low Precision Switch to Quality mode in Settings to use the Gemini LLM for deeper analysis.

Versioning & Updates

Intellistream uses snapshot versioning in the format YYwWWx (e.g., 26w13a). The application automatically checks for updates on launch and will notify you if a newer version is available on the proxy. We are working on an auto-update feature for the client to prevent having to re-download the client.