ANALYSIS The convergence of Apple's M6 and M5 Ultra desktops, Perplexity's Portable Computer agent, and JetBrains' Junie Local coding agent signals that on-device AI agents are crossing from hobbyist workaround into a distinct, commercially supported product category. The convergence is not coincidental: hardware vendors, AI startups, and developer-tools companies are each building their piece of a stack designed to keep models, data, and agentic work off the cloud.
Why it matters
The dominant narrative in enterprise AI has been cloud-first: massive data-center clusters running frontier models, billed per token. ◆ These launches collectively argue that a meaningful class of AI work — coding, knowledge tasks, file processing — can now run locally, with zero token costs and no data leaving the machine. That reframes the competitive landscape for both chip sellers and AI-application companies.
The big picture
Apple's new Mac mini starts at $899 with the M6 chip and $1,699 with the M5 Pro, each $100 above the prior M4-generation models6,9. The Mac Studio with the M5 Ultra starts at $5,499, up $20010. Apple claims the Mac mini delivers up to 4x faster AI performance and up to 2x faster graphics versus the M4 generation. The M5 Ultra uses a quad-die 3nm architecture with inter-die bandwidth of 4.4TB/s-plus, up to a 36-core CPU, an 80-core GPU, and up to 512GB of unified memory7. The M6 is Apple's first 2nm chip for Macs, with a 12-core CPU and GPU and up to 32GB of unified memory1,8. CNBC noted that desktop computers have become "strategically important for Apple as a foothold in the AI development world" and that developers "using software like OpenClaw to create their own agents often prefer to run them on a dedicated Mac Mini computer"14.
Perplexity's Portable Computer, developed in partnership with Nvidia, runs the company's agentic Computer platform entirely on local hardware11. At launch it supports Nvidia's DGX Spark desktop and Linux machines with Nvidia RTX GPUs carrying at least 24GB of VRAM4,5. A DGX Spark costs $4,800; even an older RTX 3090 with 24GB of VRAM currently runs well over $1,500. Users can choose between Qwen 3.8 27B and PPLX 27B models. "We've basically brought the exact same UI to a fully local app," said Nate Kupp, Perplexity's vice president of engineering for infrastructure and enterprise. Work completed locally consumes no billing credits, and the system asks permission before sending any step to a cloud-hosted frontier model. Portable Computer is not yet available for Mac users; Windows support is scheduled for September.
JetBrains released Junie Local, a free version of its coding agent that runs entirely on a developer's Mac13. The tool is designed to eliminate the assembly burden that has kept local agents in expert-only territory: choosing a model, configuring quantization, and tuning context settings.
ANALYSIS Perplexity's benchmark disclosures hint at why the company rebuilt its agent harness rather than simply swapping in a smaller model. On its internal 53-task Local Knowledge Work Bench, Computer running Qwen 3.8 27B on a DGX Spark scored 82.6%, compared with 77.6% for the open-source Pi harness and 74.0% for Hermes, both using the identical model. The gap between 74.0% and 82.6% on the same hardware and base model suggests the orchestration layer — sandboxing, tool routing, local file access — matters as much as raw model quality for on-device agents.
Nvidia's role is telling. "Local AI reached an inflection point," said Nader, Nvidia's director of developer technology. ◆ For a company whose revenue story centers on data-center GPU sales, endorsing local inference signals confidence that edge hardware sales can complement, not cannibalize, cloud GPU demand. Perplexity is "very focused right now on Nvidia hardware" across DGX and RTX, while considering other platforms.
Security architecture also differentiates these products. Portable Computer uses an OS-enforced sandbox; if the sandbox is unavailable, the harness disables itself rather than running tools unprotected. ◆ That design choice positions local agents not just as cost savers but as a security upgrade over cloud-based alternatives that must transmit code and files off-device.
Apple's price increases land in a market where supply constraints loom. The Wall Street Journal flagged that "between the AI agent craze and the global memory-chip shortage, Apple might struggle to keep these refreshed little powerhouse Macs in stock".
What's next
Perplexity's Windows support for Portable Computer is scheduled for September. Apple's September iPhone event looms12, likely bringing further AI-silicon messaging. ◆ The immediate question is whether Perplexity and JetBrains extend local-agent support to Apple silicon — a move that would close the loop between the hardware Apple is shipping and the software agents that justify its AI-performance claims. Until then, the local-agent category remains split across two hardware ecosystems with no single stack spanning both.