Proactive edge AI agents, optimized for any device.

Liquid builds capable, trustworthy agents that understand your intent and take action locally — without runaway cloud costs or sending personal context to someone else's server.

The challenge with building on-device AI agents

Cloud agents rely on expensive frontier models. As agents take on longer tasks and act more often, latency and token costs compound. Running intelligence on-device makes proactive, always-available agents practical — but only when the model, harness, and hardware are designed together.

On-device agents face three core challenges:

• Devices impose hard limits.

  Most agents are built for cloud-scale compute and memory. On-device agents must reason and act within tight constraints on memory, power, and latency.

• Models and harnesses are rarely co-designed.

  An agent is only as effective as the model and harness working together. When they're designed independently, the result is more errors, wasted computation, and less reliable action.

• Cloud fallback undermines the edge.

  Every cloud handoff adds cost and latency — and sends more personal context off the device. The best edge agents keep intelligence local without sacrificing capability.

Local compute = free tokens

Model & harness, co-designed

Every LFM is customized to take full advantage of your device's capabilities and execute agentic tasks with maximum efficiency. When the harness and model understand each other, they can route work more effectively, scope up or down when necessary, and get the most out of their context budget.

Hyper personalization, only possible with local context

A co-designed agent is a well-informed agent, using on-device inference loops to constantly gather context: what's on the smartphone screen, what the car's in-cabin sensors are seeing, and what the task chain needs next.

Truly proactive agentic action

Instead of waiting for prompts and executing commands, a co-designed edge agent actively seeks ways to make itself useful. Background loops that are effectively free let agents build an understanding of user needs and perform tasks autonomously.

An edge agent that knows what you need, before you even have to ask.

Contact sales

Case studies & news

Liquid AI and Mercedes-Benz partner to scale embedded in-car intelligence

08.04Models
LFM2.5-2.6B: Deploy Agents Everywhere
AUG 4, 2026Models
LFM2.5-2.6B: Deploy Agents Everywhere

03.05Models
No Cloud, No Waiting: Tool-Calling Agents on Consumer Hardware with LFM2-24B-A2B
MAR 5, 2026Models
No Cloud, No Waiting: Tool-Calling Agents on Consumer Hardware with LFM2-24B-A2B