Plannedsip_action: n/a
Self-hosted Qwen box twin (planned)
The same twin running fully on a local GPU box with Qwen via Ollama, for buyers who cannot send data abroad.
- Buyer
- Businesses with data-residency constraints.
- Adapter needed
- Same adapters, local model source
- Time back
- Same as the twin you deploy
- Monetisation
- One-off build fee · Deployment fee plus optional support retainer
Why someone pays for this
The model source is already configurable; the packaged appliance offering is what remains planned.
Starting manifest
SKILL.md
# Self-hosted twin (planned packaging)
#
# The model source is configurable today: point the deployment at a local
# OpenAI-compatible endpoint (Qwen via Ollama). Governance, receipts and
# adapters behave identically. The packaged appliance offering is planned.
Tutorial steps
- 1Follow the self-hosting runbook and pin a local model.
- 2Run one governed task end to end with no external egress.
- 3Hand over with a receipt the client verifies themselves.
The Decision Card must show
- ·Identical governance — only the model source changes.
What the receipt proves
That governance held on local infrastructure too.
Honest limits
- Local models are usually weaker at long reasoning chains — scope the skills tighter.
- A packaged appliance offering is planned, not shipped.