The Matter Lab

To learn about El Agente Estructural, check out this thread ⬇️

bsky.app/profile/them...

To learn more about another agent released today - El Agente Quntur, check out this thread ⬇️

bsky.app/profile/them...

Kudos to the team: Changhyeok Choi, @yunhengzou.bsky.social, Marcel Müller, Han Hao, Yeonghun Kang, Juan B. Pérez-Sánchez, Ignacio Gustin, Hanyong Xu, Andrew Wang, Mohammad Ghazi Vakili, @ccrebolder.bsky.social, @aspuru.bsky.social, and @variniabernales.bsky.social.
[7/7]

This work is a step toward agentic, explainable, and human-aligned molecular construction, where the goal is not just to obtain a molecule, but to build the exact 3D structure that matches chemical intent.
[6/7]

🖐️ Execute precise XYZ-coordinate manipulations to build adducts, specific transition states, or stereochemically specific organometallic complexes that string-based methods cannot handle.
[5/7]

🧠 Apply high-level chemical logic - deducing reaction mechanisms, identifying stereochemical requirements, and translating conceptual chemistry into a coherent plan.[4/7]

By combining Multimodal LLMs with domain-specific geometry manipulation tools, the agent can:

👀 Interpret 2D reaction mechanism diagrams to understand intent and inspect molecules visually.
[3/7]

While many agentic AI systems generate or edit molecules SMILES, Estructural is designed around how human experts work: directly manipulating molecular structures by selectively modifying only the regions of interest while preserving the molecular core.
[2/7]

Our second agent, El Agente Estructural (“Structural” in Spanish) is a multimodal, natural-language–driven agent for molecular geometry generation and manipulation.

🔗 www.arxiv.org/abs/2602.048...
[1/7]

-- , Eric S. Isbrandt, Mohammad Ghazi Vakili, Hanyong Xu, @ccrebolder.bsky.social, @variniabernales.bsky.social, and @aspuru.bsky.social.

[7/7]

Kudos to the authors: Juan B. Pérez-Sánchez, @yunhengzou.bsky.social, Jorge A. Campos-Gonzalez-Angulo, Marcel Müller, Ignacio Gustin, Andrew Wang, Han Hao, Tsz Wai Ko, Changhyeok Choi --
[6/7]

🔬With El Agente Quntur, researchers can delegate high-level chemistry questions to be investigated by quantum chemistry at light speed. We can’t wait for you to try it — we’re releasing soon at elagente.ca.
[5/7]

This is achieved by granting the agent full understanding and control over ORCA (Max-Planck-Institut für Kohlenforschung) quantum chemistry software (@faccts.de), achieving remarkable handling of queries with incredible complexity and depth (see our accelerated demo below).
[4/7]

🤖 Thanks to its agentic architecture and design strategies, Quntur generalizes across a wide range of quantum-chemistry tasks - from electronic structure and spectroscopy to reaction mechanisms and thermochemistry.
[3/7]

⚛️ El Agente Quntur is a research collaborator agent for quantum chemistry, who can autonomously tackle difficult, long-term tasks with minimal human supervision.

🔗 www.arxiv.org/abs/2602.04850
[2/7]

The future of quantum chemistry research is being redefined today.

After 8 months of exploration and hard work, we are thrilled to introduce two agents for chemistry:

⚛️ El Agente Quntur for quantum computing, and

🔬 El Agente Estructural for molecular geometry generation and manipulation.
[1/7]

Decoding the Suzuki–Miyaura cross-coupling reaction with the AC’s Cher-Tian Ser

Read the interview here: www.chemistry.utoronto.ca/news/decodin...

Blog post on the Matter Blotter: aspuru.substack.com/p/bulky-liga...

[2/2]

We are proud to have @chertianser.bsky.social interviewed by the UofT Chemistry department 🎉

In addition to discussing the background to his latest paper (quoted), he also highlights the role played by our SDLs in automating the experimental and computational work presented therein.
[1/2]

Renovations are really coming along at #UofT's Lash Miller building! This updated building will be the central hub of the Acceleration Consortium and our centre for materials research and innovation! @chemuoft.bsky.social @utoronto.ca @aspuru.bsky.social

Kudos to the authors: Junru Ren, Abhijoy Mandal, Rama El-khawaldeh, Shi Xuan Leong, @profhein.bsky.social, @aspuru.bsky.social, @lanalpa.bsky.social and Kourosh Darvish.
[6/6]

This approach enables more reliable real-time monitoring for automated workflows such as liquid–liquid extraction, distillation, and crystallization—bringing us closer to truly adaptive, autonomous chemistry labs.
[5/6]

LLMs reason about the experimental process, and its output is used as contextual guidance for object detection—helping the model disambiguate visually similar substances. This approach outperforms standard computer vision models both with and without external context.
[4/6]

We extend YOLO into a multi-modal, context-aware detection framework (YOLO-text) that combines:

📷 Scene images from lab experiments
🧪 Optionally, Chemistry experimentation context provided by experimentalists.
[3/6]

In this work, we tackle a key challenge in lab automation: detecting and classifying chemical phases (e.g. solids, liquids etc.) whose appearance depends strongly on experimental context.
[2/6]

Now in Digital Discovery: Context-aware computer vision for chemical reaction state detection.

🔗 pubs.rsc.org/en/content/a...
[1/6]

AI4X–Accelerate Conference 2026 abstracts have been extended until February 9 (final extension).

Join us in Singapore, June 15–19, 2026.

Submit: ai4x.cc/call-for-sub...

Kudos to all authors: Seongmin Kim, Jaehwan Choi, Kunik Jang, Junkil Park, @variniabernales.bsky.social, @aspuru.bsky.social, Yousung Jung
[7/7]

🔎Key takeaway: multi-agent debate + tool-level grounding can make computational discovery more directly actionable for experimental realization, reducing the integration burden for non-experts.
[6/7]

🧠 In Thinking mode, it uses tool-grounded multi-agent deliberation to refine synthesis routes—cross-checking evidence from models, constraints, and literature—and to propose mechanistic hypotheses supported by physics-grounded verification.
[5/7]

⚡ In Instant mode, Materealize rapidly composes tools to produce property-conditioned, synthesizable candidates along with actionable synthesis recipes.
[4/7]