by ai-in-pm
Multi-agent demo platform for Titans (arXiv:2501.00663) — neural networks that learn to memorize at test time. 7 AI agents, native desktop UI.
# Add to your Claude Code skills
git clone https://github.com/ai-in-pm/Titans---Learning-to-Memorize-at-Test-TimeGuides for using ai agents skills like Titans---Learning-to-Memorize-at-Test-Time.
Titans---Learning-to-Memorize-at-Test-Time is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by ai-in-pm. Multi-agent demo platform for Titans (arXiv:2501.00663) — neural networks that learn to memorize at test time. 7 AI agents, native desktop UI. It has 309 GitHub stars.
Titans---Learning-to-Memorize-at-Test-Time's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/ai-in-pm/Titans---Learning-to-Memorize-at-Test-Time" and add it to your Claude Code skills directory (see the Installation section above).
Titans---Learning-to-Memorize-at-Test-Time is primarily written in Python. It is open-source under ai-in-pm on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh Titans---Learning-to-Memorize-at-Test-Time against similar tools.
No comments yet. Be the first to share your thoughts!
Unlocks once the catalog security scan passes (runs nightly).
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
An interactive multi-agent demonstration platform for the landmark Titans architecture — the first neural network to learn how to memorize at test time.
The Titans paper introduces a groundbreaking memory architecture that learns what to remember during inference — no more fixed context windows. This repository brings those ideas to life with:
# 1. Clone the repo
git clone https://github.com/ai-in-pm/Titans---Learning-to-Memorize-at-Test-Time.git
cd Titans---Learning-to-Memorize-at-Test-Time
# 2. Install dependencies
pip install -r requirements.txt
# 3. Configure API keys
cp .env.sample .env
# Edit .env and add your API keys (only the providers you want to use)
# 4. Launch
python main.py
Windows users: Run
titans.bat(handles path setup automatically) or launchtitans.exefor a bundled, dependency-free experience.
Each agent explores a distinct component of the Titans architecture through a different LLM lens:
| # | Agent | Provider | Titans Role |
|---|---|---|---|
| 1 | Neural Memory Module | OpenAI (GPT-4) | Core long-term memory model |
| 2 | Memory as Context | Anthropic (Claude) | Attention-based context memory |
| 3 | Memory as Gate | Mistral | Gating mechanism for memory flow |
| 4 | Memory as Layer | Groq | Per-layer memory integration |
| 5 | Experimental Validation | Google Gemini | Benchmarking & ablation analysis |
| 6 | Innovations | Cohere | Novel extensions & improvements |
| 7 | Analysis | Emergence | Cross-agent synthesis & insights |
The native Tkinter interface provides a rich interactive environment:
Copy .env.sample to .env and add the keys for any providers you want to use:
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
MISTRAL_API_KEY=...
GROQ_API_KEY=...
GOOGLE_API_KEY=...
COHERE_API_KEY=...
EMERGENCE_API_KEY=...
You do not need all keys — the platform works with any subset and shows a graceful status for unavailable agents.
The Titans paper proposes three distinct ways to integrate a neural long-term memory module into transformer models:
The key innovation is test-time learning of what to memorize: the memory module updates its parameters during inference based on a surprise metric, allowing the model to adaptively retain information that contradicts its current knowledge — without any additional training.
Titans---Learning-to-Memorize-at-Test-Time/
├── main.py # Desktop application entry point
├── titans.bat # Windows launcher (handles path setup automatically)
├── titans.exe # Pre-built Windows executable (no Python required)
├── requirements.txt # Python dependencies
├── .env.sample # API key template
├── agents/ # Provider-specific agent implementations
│ ├── openai_agent.py
│ ├── anthropic_agent.py
│ ├── mistral_agent.py
│ ├── groq_agent.py
│ ├── gemini_agent.py
│ ├── cohere_agent.py
│ └── emergence_agent.py
├── static/ # UI assets
└── Titans Paper.pdf # The original research paper (arXiv:2501.00663)
| Problem | Solution |
|---|---|
| App closes immediately on launch | Run via titans.bat to read the terminal error output |
python main.py fails with path error |
cd into the project folder first |
google.generativeai deprecation warnings |
Non-fatal — the app still works correctly |
| An agent shows "unavailable" | That provider's API key is missing or invalid in .env |
If this project helps your research or learning, please cite the original paper:
@article{behrouz2025titans,
title = {Titans: Learning to Memorize at Test Time},
author = {Ali Behrouz and Peilin Zhong and Vahab Mirrokni},
journal = {arXiv preprint arXiv:2501.00663},
year = {2025},
url = {https://arxiv.org/abs/2501.00663}
}
Contributions are warmly welcome! Here's how to get involved:
Distributed under the MIT License. See LICENSE for full details.
Made with ❤️ by ai-in-pm · Inspired by the Titans paper
Found this useful? Please give it a ⭐ — it really helps!