Liquid AI Releases Tool for Private, On-Premise Agents
What happened
Liquid AI has released "LocalCowork," a new tool that allows agentic workflows to run entirely on-premises. The move is part of a growing trend toward local, secure, and privacy-first AI solutions for companies handling sensitive user data.
Why it matters
The underlying model, LFM2-24B-A2B, is a sparse Mixture-of-Experts (MoE) architecture with 24 billion total parameters, but only 2 billion are active per token. This design allows it to run efficiently on consumer-grade hardware, fitting within 32GB of RAM. This on-device capability is a core part of a larger trend toward privacy-first AI, as enterprises in regulated fields like finance and healthcare seek to leverage AI without sending sensitive data to third-party clouds. Solutions like LocalCowork aim to prevent data egress, making them suitable for environments with strict compliance and data sovereignty requirements. LocalCowork uses an open standard called the Model Context Protocol (MCP) to interact with local files and applications. Originally introduced by Anthropic in late 2024, MCP acts like a universal connector, allowing AI models to securely access and use external tools and data sources without custom integrations. The desktop agent ships with 75 tools across 14 MCP servers, enabling it to perform tasks like searching filesystems, running Optical Character Recognition (OCR), and scanning for leaked API keys or personally identifiable information (PII) directly on a user's machine. Liquid AI is a spin-off from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), founded by Ramin Hasani, Mathias Lechner, Alexander Amini, and Daniela Rus. The company is focused on building efficient, non-transformer-based foundation models. The company has raised significant capital, including a $297 million Series A, attracting investment from firms like AMD and OSS Capital. This funding positions them as a key player in the development of alternative, efficient AI architectures.
Key numbers
- The underlying model, LFM2-24B-A2B, is a sparse Mixture-of-Experts (MoE) architecture with 24 billion total parameters, but only 2 billion are active per token.
- This design allows it to run efficiently on consumer-grade hardware, fitting within 32GB of RAM.
- Originally introduced by Anthropic in late 2024, MCP acts like a universal connector, allowing AI models to securely access and use external tools and data sources without custom integrations.
- The company has raised significant capital, including a $297 million Series A, attracting investment from firms like AMD and OSS Capital.
What happens next
- Solutions like LocalCowork aim to prevent data egress, making them suitable for environments with strict compliance and data sovereignty requirements.
Quick answers
What happened in Liquid AI Releases Tool for Private, On-Premise Agents?
Liquid AI has released "LocalCowork," a new tool that allows agentic workflows to run entirely on-premises. The move is part of a growing trend toward local, secure, and privacy-first AI solutions for companies handling sensitive user data.
Why does Liquid AI Releases Tool for Private, On-Premise Agents matter?
The underlying model, LFM2-24B-A2B, is a sparse Mixture-of-Experts (MoE) architecture with 24 billion total parameters, but only 2 billion are active per token. This design allows it to run efficiently on consumer-grade hardware, fitting within 32GB of RAM. This on-device capability is a core part of a larger trend toward privacy-first AI, as enterprises in regulated fields like finance and healthcare seek to leverage AI without sending sensitive data to third-party clouds. Solutions like LocalCowork aim to prevent data egress, making them suitable for environments with strict compliance and data sovereignty requirements. LocalCowork uses an open standard called the Model Context Protocol (MCP) to interact with local files and applications. Originally introduced by Anthropic in late 2024, MCP acts like a universal connector, allowing AI models to securely access and use external tools and data sources without custom integrations. The desktop agent ships with 75 tools across 14 MCP servers, enabling it to perform tasks like searching filesystems, running Optical Character Recognition (OCR), and scanning for leaked API keys or personally identifiable information (PII) directly on a user's machine. Liquid AI is a spin-off from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), founded by Ramin Hasani, Mathias Lechner, Alexander Amini, and Daniela Rus. The company is focused on building efficient, non-transformer-based foundation models. The company has raised significant capital, including a $297 million Series A, attracting investment from firms like AMD and OSS Capital. This funding positions them as a key player in the development of alternative, efficient AI architectures.
