Skip to content

embernet/thewall

Repository files navigation

The Wall

AI-powered intelligence surface for meetings, research & thinking

The Wall is a desktop application that acts as your real-time AI companion during meetings, research sessions, and collaborative thinking. It listens, transcribes, analyses, and generates structured insight cards across a configurable column-based workspace — so you can think deeper, challenge assumptions, and create meaningful progress.

Free to use under the MIT Licence


Features

Three Operating Modes

Each mode controls how aggressively the AI agents respond to your content:

Mode Description
Silent Agents only run when explicitly triggered via Chat. No automatic dispatch on transcript — ideal for focused note-taking where you want full control.
Active Agents auto-run on transcript with fast dispatch (3 second debounce). The most proactive mode — best for meetings or live brainstorming.
Sidekick Agents auto-run with a gentler pace (6 second debounce). Optimised for 1:1 collaboration and research sessions.

Switch modes at any time from the top bar. Switching to Silent immediately stops the auto-dispatch pipeline.

Session Templates

Start sessions pre-configured for specific use cases. Each template defines which agents and columns are active, the operating mode, and a guiding system prompt for the AI:

Template Focus
Brainstorming Idea generation, pattern finding, and creative exploration
Research Deep research, claim verification, and evidence gathering
Decision Making Trade-off analysis, risk assessment, and structured evaluation
Retrospective Pattern recognition, gap analysis, and actionable improvements
Interview Question generation, claim tracking, and insight extraction
Strategy Visionary thinking, constraint mapping, and strategic planning

Each template includes a Session Goal input where you can provide specific context (e.g. "We're evaluating whether to migrate from AWS to GCP") that guides both the AI agents and the Chat.

Custom Templates

Create your own templates with full control over:

  • Which agents are enabled for the session
  • Which columns are visible
  • Default operating mode (silent / active / sidekick)
  • System prompt that guides the AI's behaviour
  • Goal placeholder text

Custom templates are saved to your local database and appear alongside the built-in templates in the launcher.

Multi-Column Workspace

Organise your session across 15+ specialised columns:

  • Transcript — Live transcription with speaker diarisation, audio recording, card numbering, splitting, and renumbering
  • Notes — Manual and agent-generated notes
  • Concepts — Key themes and ideas extracted from the conversation
  • Ideas — Cross-referenced, actionable ideas generated from all other cards
  • Questions — Probing questions and clarifying queries
  • Claims — Assertions and fact-checked claims
  • Gaps & Risks — Missing information, assumptions, and failure modes
  • Alternatives — Alternative approaches and options
  • Actions — Extracted action items and decisions
  • Highlights — Cards flagged by you or the AI
  • Context — Reference documents and background material
  • Summary — Auto-generated per-column summaries with configurable prompts and change tracking
  • Deep Research — Researcher agent output
  • Knowledge Graph — Visual force-directed graph of concepts and relationships
  • Agent Queue — Live view of running and queued AI agent tasks
  • Trash — Deleted cards with restore capability

Each column header shows a filtered/total card count badge when search or agent filters are active. Columns include Copy and JSON buttons for quick export of their contents.

Cards

Cards are numbered sequentially within their column for easy reference. Each card displays the agent name at the top (for agent-generated cards) along with source links, timestamps, and speaker attribution.

Card actions include copy, inline edit (Markdown), view prompt, highlight, find related, link, split, renumber, and delete.

AI Agent System

35 built-in agents run autonomously in the background:

Agent Purpose
Concept Extractor Extracts key concepts, ideas, and themes
Claim Identifier Identifies assertions and claims
Claim Verifier Fact-checks identified claims
Claim Challenger Generates counter-arguments
Gap Finder Surfaces missing information and blind spots
Clarity Seeker Flags ambiguous language and suggests clarifying questions
Problem Finder Identifies potential problems and failure modes
Tension Finder Detects conflicting statements and priorities
Solution Finder Proposes solutions to identified problems
Researcher Deep research and web retrieval
Requirement Finder Extracts explicit and implicit requirements
Constraint Finder Identifies constraints and limitations
Alternative Finder Generates alternative approaches
Trade-off Enumerator Maps trade-offs between options
Pattern Finder Identifies recurring themes and structures
Cliche Finder Flags jargon and vague language
Planner Creates structured plans and action items
Refiner Improves and sharpens existing cards
Summariser Generates concise summaries
Challenger Devil's advocate against consensus
Rhetoric Generator Persuasive arguments and talking points
Collaborator Synthesis points and areas of agreement
Skeptic Raises doubts and demands evidence
Supporter Identifies strengths and reinforces valuable ideas
Knowledge Manager Curates the knowledge graph
Questioner Probing questions to deepen understanding
Chain of Thought Reasoner Step-by-step complex problem working
Problem Solver Structured problem-solving frameworks
Coach Guidance and Socratic questioning
Visionary Extrapolates future implications
Pragmatist Grounds ideas in practical reality
Idea Generator Cross-references all agent output to generate actionable ideas
Action Tracker Extracts action items and decisions
Thinker Deep reflection on topics
Image Generator Generates images from concepts

Agent-Column Matrix

The Agent Configuration panel includes a Column Matrix tab — a grid view showing which column each agent outputs to. Click any cell to reassign an agent's output column. When a session is active with template-based constraints, use the All / Session Enabled filter toggle to show only the agents and columns active in the current session.

Per-Column Agent Filtering

Each column header includes an agent filter button that lets you filter which agents' cards are shown within that column. The column badge shows filtered vs total card counts.

Knowledge Graph

An auto-populated force-directed graph visualisation of concepts, entities, topics, and claims — with relationships like "Supports", "Contradicts", "Depends On", and more. Click any node to see the cards that reference it, or double-click to trigger deep research.

The graph toolbar provides controls for:

  • Display modes — switch between force-directed, radial, and hierarchical layouts
  • Labels toggle — show or hide node labels
  • Physics toggle — freeze or animate the simulation
  • Node spacing — adjust the repulsion force between nodes

Meeting Simulation

Generate a realistic AI-simulated meeting with configurable participants, roles, personas, and context — useful for testing workflows or exploring ideas.

Chat / Inquiry

A built-in RAG (Retrieval-Augmented Generation) chat panel that searches across all your session cards to answer questions with full context. When a session template provides a system prompt or session goal, the Chat incorporates this context into its responses. Chat history is persisted per session with per-message controls.

Embeddings & Semantic Search

  • OpenAI text-embedding-3-small for neural embeddings (requires API key)
  • Local TF-IDF fallback for offline use
  • Semantic card search across all sessions (Cmd+K)

Export Formats

Export your session data in multiple formats:

  • JSON — Full structured session backup
  • Markdown — Readable document format
  • CSV — Spreadsheet-compatible
  • HTML — Styled web page
  • Obsidian — Wiki-link format for Obsidian vaults
  • Clipboard — Quick copy

Agent names appear at the front of exported card text for clear attribution.

Transcription

  • Live audio recording with real-time transcription via Web Speech API
  • OpenAI Whisper API integration for high-accuracy transcription
  • Speaker diarisation — assign different colours to different speakers, with persistent speaker management
  • Type-to-transcribe — manually type transcript segments with speaker tagging
  • Card splitting — split long transcript cards at natural breakpoints
  • Renumbering — renumber transcript cards after edits or splits

Appearance

  • Light and dark themes — switch between light and dark mode
  • SVG icons — crisp iconography throughout the interface
  • Status bar — persistent footer with session info and save status

Persistence

  • All data stored locally in SQLite via better-sqlite3
  • Auto-save every session change
  • Import/export individual sessions or full backups
  • No data ever leaves your machine unless you configure cloud APIs

Tech Stack

  • Electron — Desktop runtime
  • React 18 + TypeScript — UI
  • Vite — Build tool
  • Tailwind CSS — Styling
  • Zustand + zundo — State management with undo/redo
  • SQLite (better-sqlite3) — Local persistence
  • graphology — In-memory knowledge graph
  • @tanstack/virtual — Virtualised rendering for large card lists
  • mitt — Event bus

Getting Started

Prerequisites

  • Node.js 18+
  • npm

Install & Run

git clone https://github.com/embernet/thewall.git
cd thewall
npm install
npm run electron:dev

Build for Distribution

npm run electron:build

Configure API Keys

Open the app, start a session, then click ⚙️ SettingsAPI Keys to configure:

Slot Provider Options Purpose
Chat Anthropic Claude, OpenAI, Google Gemini AI agents and chat
Embeddings OpenAI Semantic search
Transcription OpenAI Whisper Audio transcription
Image Generation OpenAI DALL-E, Flux Image creation

All API keys are stored encrypted in your local SQLite database. Available models are fetched dynamically from each provider's API.


Help

Click ? Help in the toolbar within the app for detailed feature documentation, or see the in-app Help page.


Licence

MIT — see LICENSE


Author

Created by Mark Burnett © 2026

About

An AI wall of agents to question, challenge, fact check, analyse, summarise, with voice transcription, chat, image generation, for simulating, participating, or minuting meetings, 1-1 brainstorming, with ability to add custom agents and workflows to automate creativity and analysis and generate concise summaries and reports to drive progress.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages