Why Portable AI Matters to Your Business

AI is moving beyond chatbots that answer questions. The latest development is an AI agent that can review documents, analyse data, use business tools and complete multi-step work directly on your own computer. For a Malaysian SME, this matters because your business may already handle sensitive customer records, supplier documents, payroll files, sales reports and internal procedures that you would rather not send to a cloud service.

Perplexity has partnered with Nvidia to launch Portable Computer, a local version of its Computer agent platform. According to VentureBeat, the system is designed to run on Nvidia DGX Spark desktop systems and Linux computers with suitable Nvidia RTX graphics cards. The important idea is simple: the model, files and AI work can remain on your device, while the system can still request permission before sending selected tasks to a more powerful cloud model.

What Happened

Portable Computer packages several components that businesses normally need to configure separately. These include local AI models, an agent harness, an inference engine, business tool connectors, a security sandbox and access to applications. The platform is intended to make local AI easier to operate rather than requiring a technical team to download model files, run an inference server and connect every tool manually.

The system can work with documents, spreadsheets, web research and business applications. In one demonstration reported by VentureBeat, an agent reviewed a folder containing 1099s and investment documents, using a Qwen model with 27 billion parameters on an Nvidia DGX Spark. Another demonstration involved analysing a CSV file locally before sending the completed analysis to Slack through a connector.

Portable Computer also supports connections to Google Drive, Gmail and GitHub, while cloud escalation remains available when a local model cannot complete a task. At launch, the platform supports Qwen 3.8 27B and PPLX 27B, with Nvidia Nemotron 3.5 Lightning planned for a later release, according to the same VentureBeat report. Linux availability comes first, with Windows support expected to follow in September 2026.

Why This Matters for Malaysian SMEs

For you, local AI could be useful wherever information is sensitive, repetitive or spread across many files. A property agency could ask an agent to organise tenancy documents, identify missing information and prepare a checklist for a staff member. An accounting practice could use it to classify files, compare records and highlight unusual entries for human review. A manufacturer could analyse production spreadsheets and draft a quality report without immediately uploading every internal document to a third-party platform.

Local processing may also suit Malaysian businesses that work with customer identity information, employee records, contracts or confidential quotations. Malaysia’s Personal Data Protection Act 2010 places responsibilities on organisations handling personal data, so keeping information on controlled business devices may support a stronger data-handling process. However, local execution is not automatically compliant. You still need proper access controls, retention rules, backups, audit records and staff procedures. For regulatory guidance, refer to the Personal Data Protection Commissioner.

Another practical use is internal knowledge management. Many SMEs have important knowledge buried in folders, emails and spreadsheets. You could ask a local agent to find a previous quotation, compare supplier terms, summarise a standard operating procedure or draft a response based on approved company documents. This can reduce the time staff spend searching, while keeping the final decision with you or your team.

Business need	Possible local AI task	Human control required
Document administration	Sort, summarise and compare files	Check accuracy before filing or sending
Sales reporting	Analyse CSV exports and identify trends	Confirm figures and business assumptions
Customer service	Draft replies using approved information	Review tone, promises and personal data
Operations	Generate checklists and flag missing steps	Keep managers responsible for decisions

Local Does Not Mean Automatically Safe

Running AI on your own computer can reduce data movement, but it introduces new responsibilities. A local agent may have access to files, email, repositories or business systems. If its permissions are too broad, a mistaken instruction or compromised connector could cause problems inside your organisation.

Local-first AI should be treated like a new staff member with powerful access: give it a limited role, define what it can touch and require approval for sensitive actions.

Portable Computer’s security design is therefore significant. The source article reports that the platform uses an always-on operating-system sandbox and disables the harness when the sandbox is unavailable. It also describes a smaller set of tools that can be loaded when needed, instead of exposing every connector to the model at all times. These safeguards are useful concepts even if you are not using this particular product.

Before introducing any local agent, create separate user accounts, restrict access to shared folders, require approval before sending emails or changing records, and keep an activity log. Start with copies of documents rather than live financial or operational systems. Your staff should also know when an AI-generated result must be checked by a person.

The Bigger Picture

The shift towards local AI reflects a change in how businesses use artificial intelligence. A simple chatbot produces one response at a time. An agent may spend much longer reading documents, checking results, operating tools and revising its work. The source article describes this as a growing demand for extended agent workloads, where local processing avoids sending every step to a remote model.

Perplexity’s research also argues that local models work better when their tools and instructions are designed specifically for their capabilities. In the company’s reported Local Knowledge Work Bench evaluation, its Computer harness using Qwen 3.8 27B scored 82.6%, compared with 77.6% for Pi and 74.0% for Hermes using the same model, based on results published in the VentureBeat article. These are company-reported results, so you should not treat them as independent proof of performance.

For Malaysian SMEs, the lesson is not that every company should immediately buy specialised hardware. Instead, you should watch how local and hybrid AI develop. A practical approach is to identify one contained workflow, such as summarising internal documents or preparing weekly reports. Test it using non-sensitive files, measure the quality of the output and document who approves the final result.

Cloud AI will remain useful for tasks requiring current web information, advanced reasoning or large-scale collaboration. Local AI may be more suitable for private documents, repeated analysis and workflows that run for long periods. A hybrid setup could give you both options: keep sensitive preparation on your device and send only an approved, limited task to the cloud when necessary.

What You Should Do Next

1. List three repetitive workflows involving documents, spreadsheets or internal knowledge.
2. Classify the information involved as public, internal, confidential or personal data.
3. Choose a low-risk workflow for a controlled trial.
4. Define which actions require staff approval before an AI agent can complete them.
5. Track errors, time saved and staff feedback before expanding usage.

Portable Computer is an early sign that capable AI agents are being designed to run closer to the business. For you, the opportunity is not simply faster automation. It is the ability to decide where your data is processed, which systems an agent can access and how much control your team keeps over important decisions. Start small, secure the workflow and expand only when the results are reliable.

Ready to Streamline Your Operations?

Technology moves fast. Your operations should keep up. AutoRunBiz builds AI systems that run your daily workflows — from WhatsApp order capture to accounting. Book a free 15-min ops audit →