pguso/ai-agents-from-scratch
Ai Agents From Scratch
This project is an LLM agent framework and orchestration engine designed for building autonomous agents that reason, utilize tools, and execute multi-step plans. It provides a system for implementing the ReAct pattern, which interleaves reasoning and action cycles to solve complex problems through iterative observation and self-correction.
The framework includes a tool integration layer that connects language models to external functions and APIs using structured schemas and embedding-based routing. It also features a memory management system to persist conversation history and user preferences, maintaining long-term context across sessions.
The orchestration capabilities cover multi-agent coordination, state-based conversation management, and the execution of dependency graphs for deterministic task completion. Additionally, the system supports prompt templating, provider-agnostic model abstractions, and execution auditing to track internal reasoning steps.
Features
- Agentic Reasoning Loops - Implements a ReAct-based reasoning loop that interleaves internal thought, external action, and observation to solve complex problems.
- Embedding-Based Tool Routing - Ships an embedding-based routing system to select the most relevant tools from a large catalog using cosine similarity.
- AI Agent Development - Provides the tools and environment for building specialized autonomous agents by defining personas, rules, and tool sets.
- Iterative Reasoning Workflows - Implements iterative reasoning workflows using the ReAct pattern to solve multi-step problems through self-correction.
- LLM Tooling Integrations - Connects language models to external functions and APIs to perform real-world actions beyond text generation.
- Autonomous Agent Frameworks - Ships a comprehensive framework for building autonomous agents that reason, use tools, and execute multi-step plans.
- Conversation Memory Managers - Implements a memory management system to store and retrieve long-term context and user preferences across sessions.
- Conversation State Management - Tracks interaction history and session state to maintain coherence across stateless language model calls.
- LLM Orchestrators - Provides an orchestration engine that manages multi-agent workflows, state machines, and dependency graphs.
- Memory Management Systems - Provides a persistence layer for maintaining historical context and user preferences across extended AI interaction sessions.
- Model Provider Abstractions - Provides a common interface to normalize API interactions and decouple application logic from specific AI service providers.
- AI Tool Schemas - Uses declarative function schemas to define the arguments and return types required for models to trigger external tools.
- ReAct Pattern Implementations - Implements the ReAct pattern to interleave reasoning and action cycles for iterative problem solving and self-correction.
- Execution Audits - Tracks and logs internal reasoning steps to create explainable and debuggable AI execution audits.
- Multi-Agent Orchestration - Coordinates multiple specialized agents to work in parallel or sequence to decompose and solve complex tasks.
- Prompt Templates - Provides systems for defining and managing reusable prompt structures to guide language model behavior.
- Directed Acyclic Graph Engines - Implements logic execution by chaining modular reasoning nodes in a dependency-ordered directed acyclic graph.
Star history
How this analysis was created: This summary and feature list were written by an AI model that read the project's README and public documentation pages. Each feature links to the documentation it came from; stars, license and language come straight from the GitHub API. The model does not read the source code, and the analysis is refreshed when the project is re-analysed. Learn more on our About page.
