Using S7 SkyQUBi
Release status (as of 2026-04-14): S7 SkyQUBi is in the pre-launch testing window. Full public launch on July 7, 2026 · 7:00 AM Central. The CLI and HTTP API interactions below are stable for the current deployment. The persona chat interface (Carli / Elias / Samuel) lands with the GO-LIVE ceremony after voice calibration is completed and signed by the household’s Chief of Covenant. See README.md for the full release status.
Once you have the pod running (see INSTALL.md), there are three ways to interact with the witness network: the CLI, the HTTP API, and the upcoming chat interface.
The fastest test — ask the witnesses a question
curl -s http://127.0.0.1:57088/witness -H 'Content-Type: application/json' \
-d '{"question": "What is 2 + 2?"}' | jq
You should see something like:
{
"classification": "FERTILE",
"confidence": 0.94,
"answer": "4",
"witnesses": [
{"model": "qwen2.5:3b", "response": "4"},
{"model": "deepseek-coder:1.3b", "response": "4"},
{"model": "qwen3:0.6b", "response": "4"}
]
}
All three witnesses agreed. Classification is FERTILE. You get an answer.
Tripping the circuit breaker
Ask a question the witnesses will genuinely disagree on:
curl -s http://127.0.0.1:57088/witness -H 'Content-Type: application/json' \
-d '{"question": "Is 1 a prime number?"}' | jq
You’ll often see:
{
"classification": "BABEL",
"confidence": 0.42,
"answer": null,
"reason": "witnesses below 70% agreement threshold — circuit breaker tripped"
}
No answer. That’s the covenant working. S7 refuses rather than guesses.
Classification states
| State | Meaning |
|---|---|
| FERTILE | Witnesses agree above threshold → answer is returned |
| AMBIGUOUS | Partial agreement → returned with confidence score |
| BABEL | Too much disagreement → circuit breaker, no answer |
| UNVERIFIED | Not enough witnesses responded in time → no answer |
Reading the witnesses list
curl -s http://127.0.0.1:57088/witnesses | jq
Returns the live witness roster — which models are loaded, what cognitive plane each sits on, current health. This is what’s rendered in the live OCTi network visualization on skyqubi.com.
Full witness set (advanced)
The default install uses the lite 3+1 witness set. For the full 7+1 set (LLaMA 3.2, Mistral, Gemma 2 9B, Phi-4, Qwen 32B, DeepSeek R1, BLOOM + CWS reporter) see the deploy guide in the repo.
Each full witness sits on one cognitive plane:
| Plane | Model |
|---|---|
| Sensory | LLaMA 3.2 3B |
| Episodic | Mistral 7B |
| Semantic | Gemma 2 9B |
| Associative | Phi-4 |
| Abstract | Qwen 32B |
| Causal | DeepSeek R1 |
| Generative | BLOOM |
The CWS reporter sits at the center and refuses to speak when the seven disagree.
The chat interface
Coming July 7, 2026 at skyqubi.ai. Today it’s a scripted preview of what the real chat will look like. The underlying witness network is the same — the chat just wraps it in a friendlier interface.
What NOT to do
- Don’t expose port 57088 directly to the internet. Put a reverse proxy in front with TLS + authentication.
- Don’t assume FERTILE means “true”. It means “the witnesses agree.” Witnesses can agree on something incorrect. The covenant is a disagreement detector, not a truth oracle.
- Don’t ignore BABEL states. When the circuit breaker trips, there’s usually a real reason — research the disagreement rather than forcing an answer.
The motto
Love is the architecture. The witnesses watch each other. The covenant holds them all.