Aegeantic Framework

A production-ready framework for building autonomous LLM agents with tool execution, pattern extraction, and multi-agent orchestration

Get Started API Reference

Overview

Aegeantic is a comprehensive framework for building stateful, autonomous LLM agents. It provides a complete infrastructure for agent execution, including context management, tool execution, pattern extraction, conditional logic flows, multi-agent coordination, and DAG-based graph orchestration.

Key Philosophy: Aegeantic separates agent concerns into discrete, composable systems. Context management, pattern extraction, tool execution, and LLM providers are all independent components that work together through well-defined interfaces.

Key Features

Versioned Context

RocksDB-backed context with automatic versioning and iteration tracking. Full history of all agent state changes.

Pattern Extraction

Flexible pattern system for extracting structured data (tools, reasoning, responses) from LLM output with streaming support.

Tool Execution

Multi-mode tool execution (thread, process, async) with validation, timeout handling, and concurrent execution support.

Logic Flows

Conditional execution with stop conditions, loop-until patterns, and context health monitoring.

Event Streaming

Real-time event stream for LLM chunks, tool execution, pattern detection, and status changes.

Multi-Agent

Built-in patterns for agent chains, supervisor-worker, parallel execution, and debate-style coordination.

Resilience

Universal retry logic with exponential backoff and token bucket rate limiting for any async operation.

Validation

Extensible validation system supporting JSON Schema, custom validators, and built-in simple validator.

Graph Orchestration

DAG-based workflow execution with dynamic scheduling, failure strategies and visualization utilities.

Architecture Highlights

Provider Agnostic

Aegeantic works with any LLM provider through the LLMProvider protocol. Implement generate() and optionally stream() to integrate your provider.

Storage Abstraction

Context is backed by RocksDB for production or in-memory storage for testing. Both implement the same interface, making it easy to switch between them.

Streaming First

All operations support streaming. Events flow through AsyncIterator interfaces, enabling real-time monitoring and reactive UIs.

Type Safety

Built with type hints throughout. All public APIs are fully typed for excellent IDE support and static analysis.

Use Cases

Research Assistants: Agents that iterate through research tasks, using tools to gather information and reasoning patterns to document findings
Code Analysis: Multi-step code review agents with tool access to linters, formatters, and test runners
Data Processing: Agents that process datasets through multiple transformations with intermediate context tracking
Multi-Agent Systems: Supervisor agents coordinating specialized worker agents for complex workflows
DAG Workflows: Complex data pipelines with parallel processing, dependency management, and failure handling using GraphRunner
Interactive Applications: Streaming agent responses to build responsive chat interfaces with real-time tool execution feedback

Quick Example

from agentic import (
    Agent, AgentConfig, AgentRunner,
    ContextManager, IterationManager,
    PatternRegistry, ToolRegistry,
    RocksDBStorage, StorageConfig,
    create_default_pattern_set
)

# Initialize storage and context
storage = RocksDBStorage(StorageConfig())
storage.initialize()

iteration_mgr = IterationManager(storage)
context = ContextManager(storage, iteration_mgr)

# Setup patterns and tools
patterns = PatternRegistry(storage)
patterns.register_pattern_set(create_default_pattern_set())

tools = ToolRegistry()

# Create agent
config = AgentConfig(
    agent_id="research_agent",
    tools_allowed=["search", "summarize"],
    pattern_set="default"
)

agent = Agent(config, context, patterns, tools, your_llm_provider)
runner = AgentRunner(agent)

# Execute
result = runner.step("Research quantum computing applications")
print(result.segments.response)

Next Steps

Quick Start Guide - Get up and running in 5 minutes
Core Concepts - Understand the framework architecture
Agent System - Deep dive into agent configuration and execution
API Reference - Complete API documentation