Research Note: A Credible Offline AI Terminal Needs a BOM, Not a Vibe
Published July 2026
Question
If Pockot were built today as a reference terminal, what public component baselines would keep the estimate honest? The answer should be a bill-of-materials scaffold, not a product promise.
Source-Backed Data Points
- NVIDIA lists the Jetson Orin Nano Super Developer Kit at $249, with 67 TOPS AI performance and a 7 W to 25 W power range in the quick specs. Source: NVIDIA Jetson Orin Nano Super.
- Raspberry Pi lists AI HAT+ 2 as available for $200, with 8 GB on-board RAM and a Hailo-10H accelerator. Source: Raspberry Pi AI HAT+ 2.
- Raspberry Pi documentation states that AI HAT+ 2 supports local LLM and VLM workloads and has 40 TOPS INT4 inference performance. Source: Raspberry Pi AI HAT documentation.
- FAA battery guidance uses 100 Wh as a key passenger-battery boundary. Source: FAA Batteries.
Reading
A BOM makes optimism pay rent. A reference device needs compute board, memory, storage, battery, display or input, enclosure, cooling, microphone or camera if used, and software maintenance. Leaving any of those out makes the estimate look smaller than the device.
Current public baselines show that local AI hardware is reachable for hobbyists, but they also expose tradeoffs. A Jetson kit has a published price and power range, but a complete field device still needs storage, battery, enclosure, and thermal design. A Raspberry Pi AI HAT+ 2 has onboard accelerator memory and a published price, but it is an add-on, not a whole terminal by itself.
The BOM should therefore be scenario-based: pocket reference, field-portable reference, and desk-dev reference. Each scenario should show official prices where available, assumption prices where not, and a date stamp.
Tool Rule
Pockot should add a `device_library` with public-price and power assumptions. Tool output should say "component scaffold, not a finished product cost" and flag missing enclosure, cooling, or measured runtime.