Moats & Intelligence · tamux docs

Structural advantage

The daemon owns terminal execution, the task queue, goal runners, memory, and the agent loop in one place. That integrated ownership is what makes higher-order capability layering possible without fragmentation.

The moat set

Moat	Tagline	Core idea
M1	The agent that knows how you think	Operator model for style, risk, timing, and attention.
M2	The agent that acts before you ask	Anticipatory pre-loading, morning briefs, and stuck-work hints.
M3	Why did you do that?	Causal execution traces, alternatives considered, and counterfactual reasoning.
M4	Skills that branch and compete	Procedural memory that evolves instead of staying static.
M5	The agent that understands your stack	Semantic environment model over code, services, infra, and conventions.
M6	The memory that knows what it knows	Deep storage with provenance, confidence, and contradiction handling.
M7	Agents that coordinate like peers	Collaboration sessions, disagreement records, and weighted voting.
M8	The agent that can prove its work	Trusted provenance and stronger audit integrity.
M9	The agent that learns without being told	Implicit feedback learning from real operator behavior.
M10	The agent that builds its own tools	Runtime tool synthesis and promotion of proven generated tools.

Why it matters publicly

This moat framing explains why tamux is not just another shell-plus-LLM product. It gives a coherent narrative for the advanced features that already exist in partial or substantial form and the architecture direction they point toward.

Documentation stance: some of these ideas are already materially implemented, while others are best described as active architecture direction. The public docs should keep that distinction honest.