Connecting to server...
Catalyst AIS

The AI That Learns, Reasons, and Shows Its Work.

Not a chatbot. A persistent reasoning engine that builds a living knowledge graph, challenges its own answers, verifies facts before generating, and gets smarter with every conversation.

Free. 3 prompts. No credit card. See the difference in one question.

Learning
Continuous
Reasoning
Visible in real time
Facts
Verified before writing
Knowledge
TAG™, not text retrieval
Illustration of Catalyst AIS live reasoning map and persistent knowledge graph
Live reasoning graph Persistent memory Pre-generation verification Every concept, prediction, and verification — visible in real time.
Applications

Three products. One primitive.

The Catalyst reasoning engine ports to any domain where historical trajectories inform present decisions. Three applications are in production today.

Hokmah

LIVE

Substrate: code repositories

Architectural memory for codebases. Hokmah treats commit history as architectural evidence, revealing dependency patterns invisible to static analysis.

hokmah.dev

Zinnawa

LIVE

Substrate: SMB operations on WhatsApp

Compositional intelligence for WhatsApp-first businesses. Zinnawa understands SMB operations as evolving graphs of promises, actions, and client rhythms.

zinnawa.com

Kefal

LIVE

Substrate: cyber defense for AI-integrated SMBs

Compositional defense for AI-integrated SMBs. Kefal models your system as a reasoning graph and detects incoherence where traditional tools see only patterns.

kefal.dev

One primitive, three substrates. The compositional reasoning engine developed at Catalyst AI Research ports to any domain where historical trajectories inform present decisions. Each port takes weeks, not years, because the engine is the same.

The Problem With Every Other AI

You keep explaining. It keeps forgetting. The answers stay safe and shallow.

01Amnesia. Every session starts from zero. You re-explain context over and over.
02Black box. A spinner, then a wall of text. No idea how it got there.
03Fragile retrieval. RAG splits your documents into text chunks and hopes the right one surfaces. It often doesn't.
04No self-challenge. The system agrees with itself. One pass, one angle, one guess.
05No learning. Use it a thousand times — it never gets better. A frozen model serving frozen answers.
Why Catalyst Is Different

Chatbots generate. Catalyst reasons.

Six capabilities no other AI system combines.

🔄
Core

Continuous Learning

Every conversation makes Catalyst smarter. The knowledge graph, concept relationships, and reasoning patterns evolve permanently. Not fine-tuning — structural learning that compounds.

📚
TAG™

Not RAG. TAG.

Upload your documents. Catalyst doesn't chunk and keyword-search them — it builds a graph of concepts, ideas, and tensions. At query time, it thinks through your knowledge.

👁️
Transparency

Visible Reasoning

Watch concepts extracted, predictions made, ideas explored, facts verified — all live, before the answer arrives. You trust the output because you watched it being built.

🔬
Multi-Angle

Self-Challenging

Exploration mode attacks your question from 3 angles. Pass 2 is forced to disagree with Pass 1. A final synthesis combines the strongest insights. Like a built-in devil's advocate.

🛡️
Integrity

Fact-First Generation

Before writing a single word, Catalyst predicts what facts the response will need and verifies them. Contested claims are flagged. False claims are blocked. Not post-hoc — pre-generation.

⚡
Execution

Code + Data + Search

Execute Python in a secure sandbox. Upload PDFs, audio, images. Toggle live web search. Voice input in any language. One interface for every modality.

Trajectory-Augmented Generation

RAG retrieves text snippets. TAG retrieves understanding.

Traditional RAG splits your documents into chunks, stores them in a vector database, and hopes the right fragment surfaces at query time. It's search dressed as intelligence.

TAG is different. Catalyst reads your documents and builds a living graph of concepts, ideas, and relationships. At query time, it doesn't search your data — it reasons through it, finding connections, challenging assumptions, and self-correcting as it goes.

Why TAG feels different

Search returns fragments. Understanding returns leverage.

Ordinary RAG

Chunk documents, rank snippets, hope the right paragraph comes back. Good for retrieval. Weak for synthesis, tension, and institutional memory.

Catalyst TAG

Build a graph of concepts, relationships, and unresolved tensions. Then reason through that graph when the question actually matters.

Visual flow of TAG: Upload, Understand, Ask, Evolve
01
Upload
PDFs, DOCX, audio, images, spreadsheets — Catalyst ingests anything your team produces.
02
Understand
Catalyst extracts concepts, ideas, and tensions — building a knowledge graph, not a text index.
03
Ask
Your question activates the graph. Catalyst reasons through your organization's knowledge, not just retrieves it.
04
Evolve
Every interaction refines the graph. Catalyst gets smarter about your domain over time.
Under the Hood

What happens between your question and the answer.

While other AI jumps straight to text generation, Catalyst runs a multi-phase reasoning pipeline — and shows you every step.

01
Analyze
Concepts extracted. Structure identified. Cognitive move classified. Before a single word is written.
02
Reason
Knowledge graph activated. Predictions generated. Analogies and tensions surfaced. Ideas explored from multiple angles.
03
Verify
Facts predicted and checked before generation. Contested claims flagged. Quality enforced across multiple passes.
04
Deliver
A grounded, verified response — with a full transparency log of everything Catalyst considered.
For Teams

Your organization's knowledge, finally usable.

Upload internal docs, reports, specs, meeting transcripts. Catalyst builds understanding that every team member can query — and that improves with each conversation.

✓Separate knowledge spaces per department or project
✓Any format: PDF, DOCX, audio recordings, images
✓Any language — auto-detected, auto-translated
For Thinkers

The AI for questions that deserve more than one angle.

Researchers, analysts, strategists, founders — anyone whose work depends on depth, not speed. Catalyst finds what you wouldn't have thought to look for.

✓Multi-angle exploration with forced self-challenge
✓Live reasoning you can follow and trust
✓Persistent context — no re-explaining
Where this lands fastest

Three use cases that make the difference obvious in minutes.

Research & due diligence

Pull themes across PDFs, interviews, audio, and notes — then challenge the first conclusion before it becomes a decision.

Internal knowledge systems

Turn company documents into shared reasoning instead of a pile of files and a search box that only half-works.

Founder strategy

Use one workspace for market notes, product logic, code checks, and multilingual thinking without losing continuity.

Built for Real Work

Everything serious teams need. Nothing they don't.

🔒
Secure
Server-side auth, encrypted sessions, sandboxed code execution, admin controls.
🌍
Multilingual
Write in any language. Catalyst reasons in English for precision, responds in yours.
🎤
Multimodal
Text, voice, files, images, live web search — every input modality in one interface.
📈
Scalable
1000+ concurrent users. Queue system. WebSocket reconnection. Disk-backed persistence.

Try Catalyst on a problem ordinary AI cannot handle.

One question is enough to feel the difference.

3 free prompts. No credit card. No commitment.

Connecting...
📚 TAG
Investor Demo Mode
Standby
Catalyst is ready for a live run.
0%
Standby
Investor Demo Mode
Catalyst is ready for a live run.
Persistent context, constraint handling, and structured reasoning will appear here in real time.
Current Mission
The response mission will appear here when planning starts.
Signal Mix
Concepts, predictions, and ideas will animate here during the run.
Concepts 0
Predictions 0
Ideas 0
Move —
The live stage rail will populate during the run.
👋 Welcome to Catalyst AIS! Upload a supported text file or start typing to explore a persistent, context-aware AI workflow.
Analyzing concepts...
Live
Starting 0%
Safe Partial Result
Waiting for extracted concepts...
Response Plan
Waiting for the response plan...
Partial results only show safe structure and extracted signals, not the hidden chain of thought.
Supported files and size limit are loading…
🛰️ Live Brief
Idle
Current Stage
Waiting for a prompt
Progress 0%
Elapsed 0s
Current Framing
Catalyst will summarize its current framing here.
Signals On The Right
Concepts, predictions, and ideas will show up here as soon as they are available.
Document Coverage
When a file is used, Catalyst will show which pages and sections actually made it into the run.
Stage Timeline
No active run yet.
🎯 Extracted Concepts
0

⚠️ Out-of-Distribution

Input not recognized

Confidence 0%
💭
No concepts extracted yet
🔮 Predictions
0
🎱
No predictions yet
💡 Ideas
0
🧠
No ideas extracted yet
⚡ Quick Actions

Catalyst AIS — Guide

Your trial

3 prompts You have 3 free prompts per session. Each message you send uses 1 prompt, regardless of mode. After 3 prompts, the session ends.

Need more? Contact us to get full access.

Standard mode

Just type your message and press Send. Catalyst analyzes your input through its cognitive pipeline: concept extraction, knowledge graph, world model, and fact-grounding. You get one deep, high-quality response.

The sidebar shows what Catalyst detected: concepts, predictions, ideas, intuitions, and world model state.

Exploration Multi-pass mode

Click the Exploration button in the header to activate. It stays on until you click it again.

In this mode, your message is analyzed 3 times from different angles, then a final synthesis combines the best insights:

  • Angle 1 — Initial analysis (standard response)
  • Angle 2 — Divergent perspective: what was missed, counter-arguments, different approach
  • Angle 3 — Connections: tensions between angles 1 & 2, emergent ideas, unexplored questions
  • Final Synthesis — A single, clean response combining the best of all 3 angles

The 3 angles collapse automatically when the synthesis arrives. Click on any angle header to expand it again.

Note: Exploration mode uses 1 prompt (not 3), but takes longer (~1-2 min). This is normal — each angle requires a full pipeline run.

Warning: If combined with SDE, expect 2-4 min per response, as each of the 3 passes is enriched by the SDE pipeline.

SDE Strategic Decomposition Engine

Click the SDE button in the header to activate. It's on by default.

The SDE enriches your prompt before Catalyst processes it. It adds:

  • Semantic Grounding — Reference definitions for key concepts from dictionaries and encyclopedias (no LLM, instant)
  • Strategic Decomposition — Multiple ways to break down your question into independent axes of investigation
  • Cross-Domain Insights — Structural isomorphisms from other fields (physics, game theory, biology...)
  • Blind Spot Detection — Hidden assumptions, missing data, unasked questions, excluded scenarios
  • Adversarial Validation — Each investigation track is stress-tested; weak ones are eliminated

The result: Catalyst doesn't just answer your question — it understands the structure of your problem before responding.

Performance impact: SDE adds ~10-15 seconds to each response (5 lightweight LLM calls). Disable it for faster responses on simple questions.

Warning: If combined with Exploration mode, expect 2-4 min per response.

Running Python code

Catalyst can execute Python code. Two ways:

  • Use /run followed by a Python code block
  • Include a ```python code block in your message and ask to "execute" or "run" it

Code runs in a secure sandbox with a 5-minute timeout. Results appear directly in the chat.

Catalyst also auto-verifies code claims in its own responses by running them.

TAG Trajectory-Augmented Generation

Your organization's knowledge, understood — not just searched.

Upload your documents, audio files, or images. Catalyst reads them and builds an internal knowledge graph: not storing text, but extracting concepts, ideas, and the relationships between them.

When you ask a question, Catalyst doesn't search for matching keywords. It thinks through your knowledge — exploring angles, finding hidden connections, challenging assumptions, and self-correcting along the way.

Traditional RAG retrieves text. TAG retrieves understanding.

Supported formats:

  • Documents: PDF, DOCX, TXT, Markdown, PPTX, XLSX, CSV
  • Audio: MP3, WAV, M4A, OGG, WebM (auto-transcribed)
  • Images: PNG, JPG (content analysis)

How to use:

  • Click the 📚 TAG button in the header
  • Create or select a knowledge space (namespace)
  • Upload your files — processing happens in the background
  • Return to chat — Catalyst now thinks with your organization's knowledge

Tip: Create separate knowledge spaces for different projects, departments, or clients.

Web Search

Click the Web Search toggle in the header to enable real-time web lookups.

When enabled, Catalyst automatically detects when your question needs fresh information from the web — for example, current events, recent news, prices, scores, or anything time-sensitive.

If your question doesn't need web data (e.g. "explain quantum entanglement"), Catalyst skips the search and responds directly, even with the toggle on.

Triggers web search: questions about latest news, current prices, today's weather, recent releases, trending topics, etc.

Skips web search: general knowledge, analysis, code help, document questions, creative writing, etc.

Tip: Leave it on — Catalyst is smart about when to use it.

Tips

  • Upload files (PDF, images, text) using the paperclip icon
  • Catalyst works in any language — it auto-detects and responds accordingly
  • The sidebar is rich with data — explore concepts, ideas, and predictions
  • Complex questions get better results: Catalyst shines on deep, multi-faceted topics

Contact

Email: [email protected]
WhatsApp: +972 522594177