docs

Pattern Documentation

Pattern is a multi-agent cognitive support system designed specifically for ADHD brains, inspired by MemGPT's stateful agent architecture.

Current State

✅ Working: Agent groups, Bluesky integration, Discord bot, data sources, CLI
🚧 In Progress: MCP client, API server, Discord slash commands
📋 Planned: Task management, MCP server

Documentation Structure

📐 Core Architecture

• Pattern Agent Architecture - Agent framework design
• Database Backend - SurrealDB patterns and entities
• Memory System - MemGPT-style memory blocks
• Tool System - Multi-operation tool architecture
• Context Building - How agent context is constructed

🔧 Implementation Guides

• Data Sources - How to integrate data sources with agents
• Group Coordination - Using agent groups and patterns
• Tool System Guide - Creating and registering tools
• Config Design - Configuration system architecture

🔌 Integration Guides

• Discord Setup - Discord bot configuration
• Bluesky Integration - ATProto firehose setup
• MCP Integration - Model Context Protocol (planned)

📚 API References

• Database API - Database operations and patterns
• Quick Reference - Common patterns and code snippets

🐛 Troubleshooting

• Known API Issues - Provider-specific workarounds
• Agent Loops - Preventing message loops
• Discord Issues - Discord integration issues
• SurrealDB Patterns - Database gotchas

Quick Start

CLI Usage

# Chat with a single agent
pattern-cli chat

# Chat with an agent group
pattern-cli chat --group main

# Use Discord integration
pattern-cli chat --discord

# Manage agent groups
pattern-cli group create MyGroup --description "Test group" --pattern round-robin
pattern-cli group add-member MyGroup agent-name --role member
pattern-cli group status MyGroup

Creating an Agent

use pattern_core::agent::DatabaseAgent;

let agent = DatabaseAgent::new(
    agent_id,
    user_id,
    agent_type,
    name,
    system_prompt,
    memory,
    db.clone(),
    model_provider,
    tool_registry,
    embedding_provider,
    heartbeat_sender,
).await?;

Adding a Data Source

use pattern_core::data_source::{FileDataSource, FileStorageMode};

let source = FileDataSource::new(
    "docs".to_string(),
    PathBuf::from("./docs"),
    FileStorageMode::Indexed,
    Some(embedding_provider),
)?;

coordinator.add_source("docs", source).await?;

Development

See CLAUDE.md in the project root for:

• Current development priorities
• Implementation guidelines
• Known issues and workarounds
• Build commands

Key Concepts

Agent Groups

Groups allow multiple agents to collaborate using coordination patterns:

• Round-robin: Fair distribution
• Dynamic: Capability-based routing
• Pipeline: Sequential processing
• Supervisor: Hierarchical delegation
• Voting: Consensus decisions
• Sleeptime: Background monitoring

Memory System

MemGPT-style memory with:

• Core memory: Always in context (persona, human)
• Archival memory: Searchable long-term storage
• Conversation history: Recent messages
• Thread-safe with Arc

Data Sources

Flexible data ingestion:

• File watching with indexing
• Discord message streams
• Bluesky firehose
• Custom sources via trait implementation