Explanation

Deepen your understanding of why Council works the way it does.

Explanation is understanding-oriented — it clarifies concepts, explores trade-offs, and provides context for design decisions. Read these to build mental models of how Council’s architecture fits together, why certain constraints exist, and how to reason about multi-expert deliberation.

Core Concepts

• Deliberation Model — How Council orchestrates multi-expert debates through structured disagreement, freeform vs. structured modes, and why multi-agent debate surfaces better insights than single-agent advice.

• Anti-Sycophancy Design — The 3-layer quality gate that prevents experts from agreeing performatively and enforces genuine, substantive disagreement.

• Memory Model — How experts remember past debates, why generic and persona experts have different memory models, and how to inspect and reset memory.

• Document RAG — How Council indexes and retrieves user documents to ground expert reasoning in real context: detection, extraction, chunking, FTS5 indexing, and recency weighting.

• Persona Experts — How document-trained experts combine LLM-synthesized profiles with RAG retrieval to reason like specific individuals, not generic roles.

Orchestration & Control

• Moderation Strategies — How Council controls turn order and prompt generation in freeform debates: round-robin, devil’s advocate, consensus-check, and when to use each.

• Context Window Management — How Council keeps long debates within model token limits through visibility scoping, rolling summaries, and hard caps.

Architecture & Security

• Architecture Overview — High-level architectural patterns, module boundaries, the 8-section system prompt structure, and the engine abstraction that makes Council provider-flexible.

• Security and Privacy — What Council stores locally, what it sends to AI providers, how secrets are handled, and the layered defenses against prompt injection in documents and cross-expert turns.