What makes it different

Existing portable-AI sticks ship a single model offline. usb.plan.ai is not another “portable Ollama setup.” It is an auditable decision and review assistant.

Council, not single-shot

User question ──▶ Stage 1: First Opinions   (parallel)
              ──▶ Stage 2: Peer Review       (anonymized ranking)
              ──▶ Stage 3: Chairman          (reasoned synthesis + trace)

The structure follows karpathy/llm-council: parallel first opinions, anonymized peer review/ranking, Chairman synthesis. This version adds a local role council for weak hardware. See the Council overview.

Compared to the inspiration

Aspect	PortableMind (commercial)	techjarves (OSS, DIY)	usb.plan.ai
License	Closed	MIT	Apache-2.0 (code)
Core	Single-model, offline	Single-model (AnythingLLM/Ollama)	Auditable Council (local role council → local multi-model council)
Council mode	No	No	First Opinion → Peer Review → Chairman
Online mode	No	No	Optional online frontier council, opt-in, Privacy-Diff
Engine	Proprietary	Ollama	llama.cpp llama-server (+ optional Ollama)
Voice / Vision	Yes	No	Yes, local (Whisper + VLM), from phase 5
Encryption	AES-256 (opt.)	-	Passphrase vault (Argon2id + AES-256)
Platforms	Win, macOS	Win, Mac, Linux	Win, macOS, Linux (+ browser-only fallback)
Price	$49-$129	Free	Free (OSS)

Three primary personas

Persona	Job to be done
Privacy Developer	Review code, logs, architecture locally; escalate online explicitly when needed.
Field Researcher / Journalist / NGO	Process sensitive notes, photos, documents on the move under changing connectivity.
Security / Compliance Power-User	Traceable model opinions, minimal-trace, BYO keys.