Readiness update, 2026-08-08: Maple AI OS is a self-hosted developer preview. The local authority loop and signed receipts are runnable. Enterprise, cooperative, and chain trust backends are not production implementations, and no Hermon domain adapter is represented here as a frontier or guaranteed production model.
We are sharpening how we describe what we build. Maple AI OS is an accountable agent runtime — the durable authority layer that sits between a model proposal and a controlled consequence. It runs above the host operating system; it does not replace the host kernel.
The core loop is unchanged and uncompromising:
LLM proposes · Guard validates · Kernel authorizes · Runtime executes · WorldLine records · Receipts prove · Evals judge · Evolution promotes or rolls back · CreditChain optionally anchors, verifies, settles, and rewards.
Local-first inference with the Hermon engine
Maple can evaluate its authority path without a cloud key. The Hermon engine is an independent inference product that Maple can route to in two ways:
- Embedded deterministic engine — a small, reproducible inference floor used for tests and offline lifecycle verification. It is not a frontier language model.
- Hermon server — a GGUF server for open-weight models, reached over local OpenAI, Anthropic, or Ollama-compatible APIs.
The router can prefer an available external engine and fall back to a deterministic planner. Model selection is deployment-specific; Maple does not designate one open model as a universal production default. The model only proposes: it drafts a typed action, Guard validates it, and the fail-closed Kernel decides whether that proposal may proceed.
Every action carries a proof
The local lifecycle produces a signed receipt and appends lifecycle evidence to WorldLine. The current developer preview verifies receipt integrity and hash-chain tampering locally. Cross-node reproduction is a separate Proof of Contribution prototype and should not be inferred from every receipt.
Proof of Contribution
Proof of Contribution is a protocol research track with a runnable local, deterministic reproduction demo. Reputation-weighted distributed quorum, adversarial operation, and economic settlement require further implementation and validation. They are design requirements, not deployed guarantees.
Blockchain-capable, not blockchain-captive
Maple boots and runs locally. The Local trust backend is the working reference path. Enterprise, Cooperative, and CreditChain modes define interfaces or prototypes and fail closed where a backend is not implemented. The default local build does not require blockchain code.
Where this is going
Models will keep commoditizing. Governed execution will not. Maple is building that layer in public, with maturity claims tied to runnable evidence rather than roadmap breadth.