Core Concepts
Core Concepts
Understand how DjinnBot works under the hood. These concepts form the foundation of everything the platform does — from agent personas to parallel swarm execution.
Services, event bus, swarm executor, and how containers fit together.
Personas, coordination, built-in tool control, and the files that define an agent.
YAML workflow definitions with steps, loops, branching, and structured output.
ClawVault, semantic search, memory scoring, 3D knowledge graphs, and consolidation.
On-demand instruction sets agents load when they need them.
External tool servers converted to native agent tools via mcpo.
Autonomous agent wake-up cycles with named routines and coordination.
Agents proactively wake each other with layered guardrails against loops and runaway execution.
Tree-sitter AST indexing, KuzuDB storage, community detection, and agent tools for codebase understanding.
Python-based tool execution that reduces context usage by 30-40%+.
Isolated Docker environments with full engineering toolboxes.
JuiceFS + RustFS distributed filesystem shared across all containers.
Git worktree isolation, persistent directories, and pipeline vs pulse strategies.
AES-256-GCM encrypted credentials with per-agent access control.
Speech-to-text via faster-whisper, text-to-speech via Fish Audio and Voicebox.