Glossary
A
Active Inference - Design principle where the system proactively reduces uncertainty rather than waiting for instructions.
Arm - Specialized module in the OctoLLM architecture responsible for domain-specific tasks (Planner, Tool Executor, Retriever, Coder, Judge, Safety Guardian).
ArmCapability - Data structure describing an arm's interface, capabilities, and resource requirements.
C
Circuit Breaker - Resilience pattern preventing cascading failures when external services are unavailable.
Coder Arm - Specialized module for code generation, debugging, and refactoring.
D
Distributed Autonomy - Design principle where arms make local decisions while the orchestrator provides global coordination.
Distributed Memory - Hybrid memory architecture with global semantic memory and local episodic stores per arm.
E
Episodic Memory - Short-term, task-specific memory stored locally in each arm (Redis-backed).
G
Global Semantic Memory - Project-wide knowledge graph stored in PostgreSQL with vector embeddings for search.
H
Hierarchical Processing - Design principle reserving expensive LLM resources for complex problems by using reflex layer and small models first.
J
Judge Arm - Specialized module for output validation and quality assurance.
M
Mixture of Experts (MoE) - Architecture pattern using multiple specialized models with a gating mechanism.
Modular Specialization - Design principle where each component excels at one thing and delegates everything else.
O
Orchestrator - Central "brain" service coordinating task decomposition and arm delegation using frontier LLMs.
P
Planner Arm - Specialized module for task decomposition and workflow generation.
Provenance Metadata - Tracking information for every artifact (arm, timestamp, command hash, data sources, tests).
R
Reflex Layer - Fast preprocessing layer for pattern matching and caching without LLM involvement.
Retriever Arm - Specialized module for knowledge base search and information retrieval.
S
Safety Guardian Arm - Specialized module for PII detection, content filtering, and safety checks.
Semantic Memory - See Global Semantic Memory.
Swarm Decision-Making - Pattern where N parallel proposals are aggregated with conflict resolution.
T
TaskContract - Core data structure representing a task with goal, constraints, budget, and acceptance criteria.
Tool Executor Arm - Specialized module for executing external commands in sandboxed environments.