Farewell Phi 3.5, making way for gpt-oss-20b? HP isn’t there yet, but a shift is underway in its approach to local AI.
Until now, its showcase was called AI Companion. This application is aimed at PC Copilot+. It offers three main features:
- A general‑purpose chatbot
- File analysis (with the ability to build up to 10 libraries of PDFs, TXT and DOCX files)
- A performance optimizer with assistance for parameter tuning and troubleshooting
HP AI Companion can run a remote GPT‑4o model (on Azure OpenAI). Or a local Phi 3.5… provided you have at least 32 GB of RAM.
Looking ahead, it seems we should expect HP IQ to take the lead. This brand preserves the chatbot aspect (Ask IQ) and the file analysis (adding PowerPoints). It adds a meeting‑note taking component (Meeting Agent), an “organized memory” based on the history of interactions (Notes & Knowledge)… and more broadly a form of orchestration.
To develop its AirDrop, HP adopts a Google‑made technology
This orchestration appears in the user interface (Visor), designed to surface the “relevant actions and controls.” It draws on a technology called NearSense. This is an implementation of the D2DI (Device‑to‑Device Infrastructure). The framework comes from Google, which uses it notably for:
- Its “inter‑device SDK,” in preview
- Transferring Meet meetings between devices
- Sharing video from Pixel smartphones to other cameras
- Instant connectivity sharing on Android (Wi‑Fi network sharing without entering a password)
At HP, NearSense will first enable file transfers between PCs; then, “in the long run” with Android devices. The technology will also allow joining meetings “in one click” on Poly equipment… provided they are managed with Poly Lens. There is talk of opening it up later to printers, content distribution, and peripheral pairing (HP has a three‑year roadmap).
“Local AI,” a little less synonymous with “PC Copilot+”
For now, HP IQ is marketed, on the front end, with a range of PCs: the EliteBook X G2 Next‑Gen (G2i and Flip G2i in Intel, G2a in AMD, G2q in Qualcomm), on which there is also a preloaded stub to facilitate later retrieval of the app.
The full list of compatible devices includes around thirty references:
- Around twenty EliteBook models (Series 6, 8, and X)
- Five ProBook 4
- Four desktop PCs (three EliteDesk and the “PC in a keyboard” EliteBoard G1a)
- Three mobile workstations ZBook (HP also promises “a selection” of fixed workstations)
- Poly conference bars X32, X52 and X72
In this sense, HP IQ goes beyond the realm of PC Copilot+, aiming more directly at GPU execution. As with AI Companion, it features a “connected” mode. But the emphasis is clearly on local use. Not with Phi 3.5, but gpt-oss-20b. It is an MoE‑type model (32 experts; 21 billion parameters, of which 3.6 billion are active) with a 128k context. It requires at least 16 GB of RAM. At launch, OpenAI claimed its performance was similar to o3-mini.
Regarding the “connected” mode, HP emphasizes—more than with AI Companion—the controls it offers to enterprises.
The deployment timeline is currently imprecise. An “early access period” is expected to begin in the spring. A “limited release” should arrive in the summer to extend the app to “more notebooks, desktops and Poly bars.” The list of compatible devices is set to expand in the autumn.
