As an early access partner for NVIDIA RTX Spark, we're looking at how local compute and Open WebUI v0.12.0 help developers and teams build agents around their own data, tools, and infrastructure.
We’re excited to be an early access partner for NVIDIA RTX Spark, NVIDIA’s Windows PC platform for local AI, creation, development and gaming. With RTX Spark laptops available starting today, we’re looking forward to what developers will build with more local AI compute on their everyday machines.
With configurations offering up to 128 GB of unified memory, RTX Spark PCs give developers more room to explore capable local models. Across the industry, the surrounding software is evolving too: agents are using models to work with files, operate tools, and complete tasks across local and cloud environments.
In Open WebUI, developers can connect a local model to project knowledge and tools, then build agents that analyze data, run code, and produce usable results on infrastructure they control.
Put local agents to work #
Start with a project you already have. With a model that supports tool calling and Open Terminal configured, an agent can inspect project files, write and run code, and return a chart, report, or script within the same conversation.
In v0.12.0, we improved how agents manage longer tasks. When enabled, context compaction now summarizes earlier context between rounds of tool use. Experimental Tool Search lets models load detailed tool definitions as needed. These changes help agents use the model’s available context more efficiently as work progresses.
Expanded Skills support also lets teams package instructions, scripts, and templates into reusable procedures, with version history to review and restore changes. For developers and enterprise teams, that provides a path from an individual experiment to shared workflows that use local models, internal data, and connected tools.
Start with two commands #
The standard Python installation remains straightforward. With Python 3.11 in an activated environment:
pip install open-webui
open-webui serveOpen localhost:8080, then connect your model server. The connected server handles inference, while your conversations, files, and tools come together in the workspace.
Keep your model choices #
More capable local hardware expands your options. Hosted models can sit alongside local connections, with the selected endpoint determining where a conversation and its context are processed. You choose the combination that fits your work.
When we wrote about NVIDIA DGX Spark last year, we focused on making local AI accessible on infrastructure people control. RTX Spark brings that opportunity to a new generation of personal computers, as agents give developers more ways to put that compute to work.
Read NVIDIA’s local AI announcement: [NVIDIA BLOG LINK].
Then start with the setup guide, bring a project of your own, and share what you build with the community.
