Local-first inference
Ollama is a development adapter today; vLLM, SGLang, TGI, MLX, and Esther-trained weights fit behind the same engine boundary.
Esther · Sovereign Intelligence
Intelligence under authority.
Esther is the first-party AI platform being built to run without commercial model providers as a foundation. Stone governs authority. Sentinel watches independently. Evidence proves what happened.
Domain independent
Mounted here at writteninstone.io/esther now, ready for esther.writteninstone.io and a future dedicated Esther domain without changing Core.
Fail-closed authority
Unknown identity, missing passport, ungranted capability, revoked credentials, or Sentinel quarantine all deny action by default.
Non-negotiable proof
Remove every external provider credential. Esther must still run.
Esther CLI → Gateway → Core → sovereign inference → Stone policy → Sentinel observation → signed evidence. If that chain breaks, sovereign operation is incomplete.