Perplexity launches Portable Computer, a fully local version of Perplexity Computer with support for Qwen3.8-27B and Perplexity's PPLX 27B on Nvidia's DGX Spark
Perplexity is launching Portable Computer today, a version of its agentic “Computer” platform that runs entirely …
VentureBeatMichael Nuñez
Context & Ripple Effects
Portable Computer is the endpoint of a deliberate arc Perplexity has been running all year: February's general-purpose digital worker routed work across 19 cloud models for Max subscribers, March's Personal Computer brought an agent onto Macs, and June added the ability to split tasks between local and cloud models to keep private data on-device. Today's release removes the cloud leg entirely — orchestrator LLM, subagent LLM, and harness all running on the machine.
The launch is also a joint play with Nvidia: Perplexity credits Nvidia with supporting the research on DGX Spark, and the two companies are framing co-designed 27B-class models (Qwen3.8-27B and Perplexity's own PPLX 27B) plus a minimal harness as the way small models actually work for agents. The pickup was unusually broad — eight syndicated outlets from The Register to Computerworld within a day — signaling that zero-token local agents are being read as more than a niche release.
First-order effects
Perplexity users with a DGX Spark or RTX Linux PC can now run the full Computer agent stack on-device at zero per-token cost, with Perplexity claiming its harness scores 82.6% on real knowledge work against open-source alternatives.
Nvidia gets a flagship software demonstration for DGX Spark — an agent vendor showcasing the box rather than a benchmark — deepening a partnership where Nvidia had reportedly (and unconfirmedly) considered hiring some Perplexity staff.
Second-order effects
Cloud-based agent subscriptions face a direct pricing comparison: when a comparable agent runs free after the hardware purchase, per-token metering becomes a visible line item competitors like OpenClaw-style rivals have to answer.
Open-weight model publishers around the 27B class gain a new distribution channel, since Perplexity's argument that models and harnesses must be designed together pushes demand toward models tuned for compact agent scaffolds.
Third-order effects
If agentic inference keeps moving to local silicon, the industry splits into co-designed hardware-plus-small-model stacks and frontier cloud fallbacks — with unconfirmed reports that Nvidia will raise system prices by at least 15% from early 2027 setting how expensive that local floor becomes.
Agent vendors' moat shifts from model access (anyone can route 19 APIs) to harness engineering and connector design, which is portable across machines and harder to meter.
The trend: Agentic computing is migrating from metered cloud inference toward co-designed small models running on local hardware, with cost and privacy — not capability alone — driving where the runtime lives.
Today we're launching Portable Computer on @NVIDIA DGX Spark. Portable Computer is a fully local version of Perplexity Computer, where the entire runtime: orchestrator LLM, subagent LLM, agent harness all run on your local hardware. No cloud dependency.
Big thanks to @nvidia for working together with us and supporting this research on DGX Spark as well as enabling an open ecosystem around cost-effective open-weight models, inference frameworks, and hardware with unified memory.
In a compute and power-constrained world, a good chunk of agentic inference needs to move to local hardware. A drastic version of that is a fully local agent runtime, where the model (orchestrator and subagents) and the harness run locally. Portable Computer from Perplexity is …
New research: Portable Computer is a local-first agent for private and cost-effective work. With an on-device 27B model, our harness scores 82.6% on real knowledge work, beating open-source harnesses Pi and Hermes. Our post-trained PPLX 27B reaches 85.4%.
2 months ago, we crashed Jensen's board meeting to show him Perplexity running locally on DGX Spark 🤣 (lol im holding architecture diagrams like he's gonna look at em during a board meeting) Local AI used to be for enthusiasts. Folks were running tiny quantized models on
The model and harness are designed together because small models fail in harnesses built for frontier models. Portable is a minimal system prompt, skills that load on demand, connectors as compact CLI tools instead of MCP servers, self-verification, and an always-on sandbox.
Meet Portable Computer, Perplexity's new local-first agent stack on NVIDIA DGX Spark. When running locally, Portable Computer offers one-click local inference setup and an optimized agentic experience for DGX Spark. Learn more and get started today: https://blogs.nvidia.com/...
not to beat a dead horse, but whenever i see this statement being made i immediately doubt the level of understanding of the person implementing the harness
Thanks to progress in open-weight models, it is now possible to run a powerful agent harness fully local. We adapted the Perplexity Computer harness (originally cloud-based) to a local-first version, orchestrated by a 27b MoE that runs locally on DGX Spark.
> “Local first” > “On device” > tested on a DGX Spark, a $5000 dedicated AI supercomputer that has no support for traditional software I hate that these terms don't mean anything anymore.