Structural discovery before context retrieval
The system should know what the document contains before it decides what to fetch in full.
Nyxdoc is architected on a fundamental principle: reading, writing, and review must each expose a visible contract. This transforms agent collaboration from probabilistic guesswork into an auditable system of record.
The system should know what the document contains before it decides what to fetch in full.
Patch-first writing keeps responsibility, conflict detection, and rollback visible.
Comments, diffs, and status need to stay attached to the document state they are about.
Our model thrives on restraint; every operation is scoped rather than rewritten. This precision is what makes automation more dependable, not less capable.
Discover structure, section metadata, and addressable blocks first. Fetch only the blocks a task actually depends on.
Changes should carry preconditions, revision boundaries, and a history that explains how the current state was produced.
Discussion is not side decoration. It becomes the bridge from human instruction to agent execution and back to review.
The task starts on a specific block, section, or document area instead of in a vague request disconnected from context.
Context loading becomes intentional, cheaper, and easier to explain back to a reviewer.
Every important change stays attributable, diffable, and tied to a known revision boundary.
Oversight becomes a native product interface, replacing the manual reconstruction of past context.
Read the philosophy behind calm interfaces, visible change, and human judgment that remains explicit.