DOCSEARCHbyAlgolia
Submit Docs

LLM Council

website·Indexed 30 days ago

https://amiable.dev/

A multi-LLM deliberation system where multiple Large Language Models collaboratively answer questions. Developed by amiable-dev, the project focuses on collaborative AI reasoning and multi-model security reviews.

Tokens
44.1K
Snippets
182
Records
240
Agent score
50%

What's inside LLM Council

  1. Overview of SmartBadge for GitHub repositories

    SmartBadge is a tool for GitHub repositories that provides AI-powered semantic analysis of a codebase. Unlike traditional repository statistics, it uses semantic intelligence to provide insights into the purpose, architecture, and quality of the code.

  2. Overview of the llm-council deliberation system

    llm-council is a multi-LLM deliberation system designed to produce higher-quality AI responses by having multiple models collaboratively answer questions through a 3-stage process: parallel responses, peer review, and chairman synthesis. It is available as a Python library (llm-council-core on PyPI), an MCP server (integrating with Claude Code), or an HTTP API (built with FastAPI).

  3. Implement a unified multi-platform chatbot architecture

    To deploy a chatbot across multiple platforms (Discord, Slack, Telegram, WhatsApp) from a single codebase, use a 'thin adapter' and 'central gateway' architecture. Platform Adapters handle platform-specific concerns (authentication, parsing, sending), while the Central Gateway manages shared business logic including message normalization, access control, rate limiting, LLM orchestration, and context management.

  4. Manage agent port conflicts with berth

    The tool berth provides a solution for port conflicts when running multiple AI agent sessions (such as Claude Code) on a single machine. It assigns every project a permanent, decodable block of ports to ensure agents and terminals agree on port usage without terminating active processes.

  5. Implement the 5-Block Architecture for AI agent memory

    The 5-Block Architecture manages finite token windows by dividing the prompt into five prioritized blocks. Each block has a specific purpose, a token budget, and a truncation rank. Rank 1 is the most important (truncated last), and Rank 5 is the least important (truncated first).

    Truncation Order: Knowledge (Rank 5) $\rightarrow$ History (Rank 4) $\rightarrow$ Task (Rank 3). System and Project blocks are never truncated.

    Budget Example (8K Model):

    • System Block (Rank 1): 500 tokens
    • Project Block (Rank 2): 1,000 tokens
    • Task Block (Rank 3): 500 tokens
    • History Block (Rank 4): 1,000 tokens
    • Knowledge Block (Rank 5): 2,000 tokens
    • User Query (Reserved): 1,000 tokens
    • Response (Reserved): 2,000 tokens
    • Safety Buffer: 192 tokens

  6. Implement a Context-Aware AI Development System with Luminescent Cluster

    The Luminescent Cluster architecture implements a three-tiered memory system for AI development:

    1. Session memory (hot context)
    2. Long-term persistent knowledge (via Pixeltable)
    3. Intelligent orchestration for token-efficient workflows.

  7. Aggregate remote blog posts in Docusaurus

    To unify blog content from distributed project repositories in Docusaurus, implement a remote blog post aggregation system by extending the ADR (Architecture Decision Record) aggregation pattern.

  8. Understand Stentorosaur data access and scope

    Stentorosaur operates as a communication layer on top of status workflows. It has access to GitHub Issues (text, comments, labels, timestamps), committed JSON data (response times, uptime history), and the GitHub Actions environment. It does NOT have access to server logs, APM tools (e.g., Datadog, New Relic), or internal system metrics (CPU, memory, database queries).

  9. Automate AI agent session memory with Skills and Git Hooks

    To eliminate manual context loading and saving in AI coding sessions, implement a system combining Agent Skills for session workflows and Git Hooks for automatic synchronization. This prevents stale context, lost continuity, and knowledge base (KB) drift by automating the ingestion of code changes and the retrieval of task state.

  10. Extract and normalize ADR status from content

    Extract the status of an Architecture Decision Record by searching for a ## Status section in the markdown content. The extracted status can then be mapped to normalized categories: proposed, accepted, or deprecated.
    const STATUS_MAPPINGS = {
  proposed: ['proposed', 'draft', 'review'],
  accepted: ['accepted', 'approved', 'active'],
  deprecated: ['deprecated', 'superseded', 'obsolete'],
};

function extractStatusFromContent(content) {
  // Try ## Status section first
  const sectionMatch = content.match(/^##\s*Status\s*\n+([^\n#]+)/mi);
  if (sectionMatch) return sectionMatch[1].trim();

  // Fall back to front matter
  // ...
}

  11. Install and explore Berth for agent sessions

    Berth is a tool designed to manage port assignments (berths) for multiple concurrent agent sessions on a single machine. It operates without a proxy, DNS, daemon, or cloud infrastructure and does not automatically rewrite application configurations or terminate processes.
    npm install @amiable-dev/berth

# Check current port status to see why Berth is needed
berth check

  12. Troubleshoot 5-Block Architecture memory issues

    Common symptoms and fixes for the 5-Block Architecture implementation:

    • Empty Knowledge block: Likely caused by retrieval failure or missing memories. Verify the embedding model and ensure memories exist.
    • Wrong memories retrieved: Likely a query-memory mismatch. Adjust retrieval scoring or implement query rewriting.
    • History too short: Occurs when the budget is exceeded and older turns are dropped. Increase the history budget or improve summarization logic.
    • Task block stale: Occurs when task context isn't updated between turns. Ensure the task block is updated whenever goals change.
    • System block truncated: This should trigger an error as the system block is Rank 1. Reduce the system prompt size to fit the budget.