Concepts

The mental model behind Nexus. Read these in order — each builds on the last.

Concept	What it is
Agents	Runtime-defined entities created from the UI: model, skills, memory policy, tools, permissions, Kubernetes resources. Repo-agnostic — repositories live on the board.
Boards	A board is a project: it owns the GitHub repositories its tasks build across and maps to a Taiga project. Nexus runs many boards in parallel off one shared agent pool.
Skills	Reusable, versioned, testable units of capability assigned to agents.
Memory	Scoped, reviewable context (global, project, agent_private, task, user_profile).
Tools	Executable capabilities an agent calls mid-run (fs.*, shell.exec, git, memory.propose), gated by permissions and a hardline firewall.
Blueprints	Templates that agents are instantiated from.
Permissions	What an agent is allowed to do, and what needs human approval.
Lifecycle	Draft → enabled → quarantined → archived, and the run lifecycle.

The principle, restated

Agents are data. Skills are data. Memory is data. Execution is isolated. Nexus Core is the only authority. Taiga is only a projection.

Every concept here is a document in MongoDB (see Data model) plus the behaviour Nexus Core applies when it assembles a run.