boost.ai launches Presis, a self-hosted language model for AI agent classification, claims 2.5x speed gain
boost.ai announced Presis on November 10, 2026, a self-hosted language model that handles classification decisions around AI agent conversations. The company says internal testing found it at least 2.5 times faster than the models it replaces.
boost.ai announced the launch of Presis on November 10, 2026, a self-hosted language model for classification decisions around AI agent conversations, as reported by Unite.AI.
What Presis does
According to boost.ai, Presis decides which part of an AI agent should respond to a user. It also determines whether a question is in scope and whether an answer is safe to show an end user. The company also says it handles classification of completed conversations.
Speed claims
boost.ai says internal testing found Presis to be at least 2.5 times faster than the models it replaces. The company claims the difference is most noticeable during peak-demand hours and in voice AI. These figures come from boost.ai's own testing, and no independent benchmark results have been published.
Deployment
Presis is self-hosted and is positioned as part of the boost.ai agent platform. Parameter count, context window, pricing and benchmark scores are not yet disclosed.
What this means
Classification steps such as routing, scope checks and safety checks sit in the path of every agent reply, so their latency adds to the response time users experience. boost.ai's claim of a speed gain during peak demand and in voice AI points at those latency-sensitive cases. The self-hosted design places the model in the customer's own environment. The 2.5x figure is a company claim from internal testing and should be treated as such until independent measurements are available.
Related Articles
OpenAI's dots agent can now connect to multiple local computers and split tasks across them
OpenAI announced on November 9 (US time) that its always-on agent tool dots can now connect to multiple local computers. Each machine can be given a separate task. The feature is available to Pro, Business Premium and Enterprise plan users in eligible regions.
Google Antigravity adds Knowledge Work interface for non-coding tasks, with the same agent underneath
Google is rolling out a Knowledge Work user experience in Antigravity, designed for research, data analysis, documents and presentations. The underlying agent is unchanged; the interface hides developer controls and uses plain-language permission prompts. Google says tens of thousands of Googlers outside software engineering use Antigravity weekly.
Qianwen Office 2.0 adds "Collaboration" so teams and Agents work in one shared space
Qianwen Office (千问办公) released version 2.0 on November 9, its largest upgrade in the three months since launch. The headline feature, "Collaboration," puts multiple people and digital employees in one shared workspace, and all new features are live in the client.
Gemini now creates and edits documents, PDFs and spreadsheets in chat, including from photos of handwritten notes
Google's latest Gemini upgrade lets users create and edit documents, PDFs, spreadsheets and more directly inside Gemini. In a TechRadar hands-on, a photo of handwritten notes became a PDF that kept the key content, but a follow-up edit added empty citation markers and dropped some detail.
Comments
Loading...