Core Concepts

Core mental model and architecture for NucleusIQ.

• Agents
• Tasks — Task.resources + rendered context (v0.7.14)
• Production architecture
• Execution modes — Gearbox + v0.7.14 harness guarantees
• Autonomous mode — why the harness changed, what improved, examples
• LLM and providers
• Models
• Providers
• Prompts
• Context Management — Transcript compaction & masking (v0.7.6+; V2 stable in v0.7.7)
• Context stack (L0–L7) — Unified glossary: L0–L3 compaction tiers, L4–L6 run-local state, L7 synthesis/AgentResult
• Run-local context state — Per-run workspace, evidence, lexical corpus, phase telemetry (v0.7.8)
• Runtime components
• Memory
• Streaming
• Structured output — result.parsed / result.schema_valid (never treat AgentResult as the schema)
• Attachments
• Plugins overview
• Error handling