Proxy Base Agent

What is Proxy Base Agent

The Proxy Base Agent (PBA) is a foundational agent framework that enables the creation of custom agents using any language model (LLM) with guaranteed state consistency and controllable execution. It is built on the Proxy Structuring Engine (PSE) and focuses on stateful interactions and extensibility.

Use cases

Use cases for the Proxy Base Agent include building intelligent chatbots, developing interactive educational tools, creating automated report generation systems, and enhancing existing applications with advanced language model capabilities.

How to use

To use the Proxy Base Agent, clone the repository from GitHub, install the necessary dependencies using pip, and launch the interactive setup wizard with Python. Ensure you have Python 3.10 or higher and a compatible operating system.

Key features

Key features of the Proxy Base Agent include a structured state machine architecture for rapid prototyping, support for local inference via Huggingface Transformers, goal-oriented interactions, multi-step reasoning, and the ability to utilize external tools.

Where to use

The Proxy Base Agent can be used in various fields such as artificial intelligence development, natural language processing, automated customer service, and any application requiring stateful interactions with language models.

A stateful agent with 100% reliable tool use.

Build custom agents on any LLM with guaranteed state consistency and 100% reliable tool use.
Powered by the Proxy Structuring Engine (PSE).

The Problem: Agents That Don’t Work

Most LLM agents rely on fragile prompt engineering. This leads to unpredictable state management, hallucinated tool arguments, and frequent execution failures – making reliable automation nearly impossible.

The Proxy Base Agent (PBA) is the engineered solution. PBA uses the Proxy Structuring Engine (PSE) – our high-performance Hierarchical State Machine (HSM) engine – to enforce reliable, stateful execution for any LLM.

With the Proxy Base Agent, you define your agent’s behavior through a defined state graph, and PSE guarantees the LLM adheres to it, step-by-step.

Key Capabilities

• Guaranteed Stateful Execution: Define agent workflows as explicit HSMs (e.g., Plan ➔ Act). PSE ensures the LLM follows the defined states and transitions precisely.
• 100% Reliable Tool Use: Eliminate runtime errors from malformed API calls or hallucinated function arguments. PSE guarantees tool calls match their required schema during generation.
• Dynamic Runtime Adaptation (MCP): Connect to external Model Context Protocol (MCP) servers on-the-fly. PBA instantly integrates new tools and capabilities with the same structural guarantees, no restarts needed.
• Model & Framework Agnostic: Run reliable agents locally using your preferred LLMs and backends (MLX, PyTorch supported).
• Modular & Extensible: Build specialized agents by adding custom tools, defining new states, or modifying the core HSM architecture.

How It Works: Reliability Through Structure

PBA’s core is an HSM enforced by PSE at runtime:

1. HSM Definition: Agent logic (states like Thinking, Tool Call) is defined as a StateMachine. Each state uses a nested PSE StateMachine to enforce its specific output structure (e.g., fenced text, JSON schema).
2. PSE Runtime Enforcement: The StructuringEngine ensures the LLM generates only valid state transitions and structurally correct output within each state. Tool call arguments are guaranteed to match the required schema.
3. Dynamic Updates (MCP): Connecting to an MCP server rebuilds the relevant parts of the HSM and reconfigures PSE instantly, making new tools reliably available.

PBA doesn’t just ask the LLM to be stateful and reliable; it engineers it through PSE’s runtime HSM governance.

Installation

pip install proxy-base-agent

(See Installation Docs for full details, development setup, and framework extras)

Quickstart

Launch the interactive setup wizard to configure your LLM and run the agent:

python -m agent

Documentation

We’ve created detailed technical documentation for the Proxy Base Agent:

• Overview
• Quickstart
• Core Concepts
• Extending the Agent
• API Reference

The Proxy Company Services

Leverage our foundational expertise to build complex, mission-critical agentic systems with PBA. We offer:

• Custom Agent Development: Tailored agents for your specific workflows.
• Advanced Integration: Connect PBA reliably to proprietary systems and APIs.
• Production Support: Architecture, performance tuning, and support contracts.

➡️ Explore Business Services & Schedule Consultation

License

Apache 2.0 (LICENSE). Depends on pse (Apache 2.0).

Citation

@software{Wind_Proxy_Base_Agent_2025,
  author       = {Wind, Jack},
  title        = {{Proxy Base Agent: Reliable AI Execution via Hierarchical State Machines}},
  version      = {1.0.0},
  date         = {2025-04-15},
  url          = {https://github.com/TheProxyCompany/proxy-base-agent},
  publisher    = {The Proxy Company},
  note         = {Leverages the Proxy Structuring Engine (PSE) for runtime guarantees}
}