Glossary
Spec — A markdown document at .aifactory/specs/<spec-id>/spec.md with description + acceptance criteria. The source of truth for a task.
Spec-id — A NNN-<slug> identifier, e.g., 001-add-healthz-endpoint. Lexicographically ordered.
Task-id — A composite <project-id>:<spec-id>. Used in API URLs.
Subtask — A line item in the implementation plan. Each has a status (pending, in_progress, done, blocked).
Worktree — A git working directory under .aifactory/worktrees/tasks/<spec-id>/. Isolated from the user's main tree.
Agent Profile — A reusable mapping of phases to model strings. Stored in ~/.aifactory/profiles.json and selectable from the Task Creation Wizard.
Phase — A discrete step in the pipeline. Spec creation has up to 7 phases; implementation has 3 (Planning, Coding, QA).
Phase Model — The model string for a given phase, e.g., sonnet, ollama:qwen3:14b, gpt-4o.
MCP — Model Context Protocol. The standard way agents discover and call external tools. AIFactory ships with Context7, Linear, Graphiti, Puppeteer MCP servers wired up by default.
Graphiti — The semantic memory layer. A knowledge graph stored in an embedded LadybugDB instance under .aifactory/specs/<spec-id>/graphiti/. Used by the planner to pull in prior-session insights.
rmux — A Rust-based fork of tmux that AIFactory bundles for the Live Agent Console. Provides byte-level pane streaming via pipe-pane.
rmux session — A tmux-style session managed by the rmux daemon. AIFactory creates one per task when AIFACTORY_RMUX_ENABLED=true.
Attach — Taking interactive control of an agent's rmux session via the browser. Exclusive — only one client at a time (409 for others).
Auto-profile — A task-level setting that lets the planner pick provider/model per phase based on inferred complexity. Overrides the user's default profile.
BMad Method — A scale-adaptive intelligence framework integrated in apps/backend/integrations/bmad/. Provides complexity detection (5 levels), three-track planning, and four named agent personas (Sarah/Winston/Alex/Jordan).
Failover — Automatic provider switching when the active provider rate-limits or errors. Order of attempts is the task's profile list, in declared order.