Install Portable Computer on DGX Spark
Sets up the local agent harness, orchestrator, two post-trained models, and sandboxed execution.
Portable Computer on NVIDIA DGX Spark
Your local-first agent stack
Portable Computer runs the agent harness, orchestrator, and post-trained models on NVIDIA DGX Spark™. Run local agents, hold state across sessions, and tap frontier cloud models when needed.
Review my 2024-2025 annual brokerage summaries, consolidated 1099s, and tax returns. Identify the two recurring holdings or behaviors creating the most avoidable fees or tax drag. Show the relevant fees, distributions, and short-term gains by year, with citations to the filename and page. Do not calculate my overall after-tax portfolio return or allocate taxes from wages.
Working · 0 steps completed
Thinking
Local agent: Statement extractionLocal Orchestrator
Fee and drag summaryLocal Orchestrator
2 subagents running
Reading 9 files in Brokerage/2024-2025/
Fees 2024-2025$9,420
Cap gain distributions$21,300
Short-term gains$6,840
Personal data secured•••-••-6642acct •••••4417
Citing every figure to a file and page
1099-DIV-2024.pdf p. 2
1099-B-2025.pdf p. 7
brokerage-summary-2025.pdf p. 4
2024-2025 Fee and Tax Drag Findings
1.The actively managed growth fund in the taxable account is the larger drag: it paid a capital gain distribution in both years, taxed the year it landed, while you added to the position (1099-DIV-2024.pdf p. 2, 1099-DIV-2025.pdf p. 2).
2.Second, short holding periods in that same account. 2025 short-term proceeds sit on lots held under a year, on top of the advisory fee tier the account is billed at (1099-B-2025.pdf p. 7, brokerage-summary-2025.pdf p. 4)…
Findings saved to Brokerage/ on this machine
Local
Cloud
GPT-5.5
Claude Fable 5
Claude Opus 4.8
Claude Sonnet 5
Hybrid
Qwen 3.8 27B Orchestrator / Claude Fable 5Own the stack your agents run on.
Each task runs locally on your DGX Spark. If a task needs frontier reasoning, the orchestrator will ask before routing a task to one of 15+ cloud models.
Your task
DGX Spark
-
Local harness and models
Portable Computer runs the agent harness, orchestrator, planner, and tool router locally with Qwen 3.8 27B or Qwen 3.6 35B. NVIDIA Nemotron 3.5 Lightning is coming soon.
-
Cloud escalation
When a step needs the web or frontier reasoning, the orchestrator asks for approval, routes that step to one of 15+ cloud models, and brings the result back to the local run.
-
Sandbox infrastructure
Code and tool execution run in isolated environments with controlled access to local files and connected apps.
-
Connectors and research
Gmail, Outlook, Slack, and GitHub work through the local orchestrator. Portable Computer can also call Perplexity Search and run wide or deep research, then return the findings to the same task.
-
Own the compute
Work handled by the local models has no per-token charge. That makes repo-scale migrations and other high-volume jobs practical to run on hardware you own.
Set up on DGX Spark.
Portable Computer brings the agent harness, orchestrator, model, and sandbox to your NVIDIA DGX Spark, with app connectors and cloud escalation in one system.
Setup steps
Download local models
Use Qwen 3.8 27B or Qwen 3.6 35B locally. NVIDIA Nemotron 3.5 Lightning is coming soon.
Start a task
Start any task locally. Stay local, upload files, connect to apps, and let Computer orchestrate the rest.
Install commands
sudo curl -fsSL https://packages.perplexity.ai/perplexity.gpg -o /usr/share/keyrings/perplexity.gpg
echo "deb [signed-by=/usr/share/keyrings/perplexity.gpg] https://packages.perplexity.ai/deb stable main" | sudo tee /etc/apt/sources.list.d/perplexity.list
sudo apt-get update
sudo apt-get install perplexity
Set up Portable Computer on NVIDIA DGX Spark.
Available today for Pro and Max subscribers.
Set up now