Get expert guidance

Technical consultations, custom engineering, and support contracts.

Proxy Base Agent

Reliable Agent Execution for Any LLM

Key Features

Stateful Execution

Define agent workflows as explicit state machines (e.g., Plan ➔ Act). The Proxy Structuring Engine ensures precise execution.

100% Reliable Tool Use

Eliminate errors from malformed API calls or hallucinated arguments. Guarantees tool calls match their schema during generation.

Dynamic Adaptation

Connect to external Model Context Protocol servers on-the-fly. Instantly integrate new tools with the same structural guarantees.

Model Agnostic

Run reliable agents locally using your preferred LLMs and backends (MLX, PyTorch supported).

Extensible Design

Build specialized agents by adding custom tools, defining new states, or modifying the core state machine architecture.

The Problem with Agent Reliability

Conventional AI agents often fail unpredictably. Relying solely on prompting leads to inconsistent state management and unreliable interactions with external tools or APIs.

The Proxy Base Agent (PBA) provides engineered reliability. It employs an explicit state machine architecture, enforced at runtime by the Proxy Structuring Engine, to guide LLM behavior. This defines agent logic (e.g., Plan ➔ Act) and ensures dependable execution.

How It Works

State Machine Logic - Agent behavior (Thinking, Tool Call states) is defined via state machines. Nested structures enforce rules within each state.

Runtime Enforcement - The Proxy Structuring Engine guarantees valid state transitions and structurally correct outputs (e.g., tool calls) during generation.

Dynamic Capabilities - Connecting to Model Context Protocol servers adds tools by rebuilding relevant state machine components at runtime.

Observable Control - Explicit states for planning and action provide transparency and allow for predictable agent behavior.

Applications

Trustworthy Automation - Build agents that reliably complete complex, multi-step tasks without unpredictable failures.

Secure Tool Interaction - Integrate safely with APIs and external systems, preventing errors from malformed requests.

Custom Agent Systems - Develop specialized agents tailored to specific domains, proprietary data, or unique workflows.

Personalized AI - Enable agents that run locally, maintain state consistently, and adapt to individual needs.

Benefits

Dependable Execution - Move from unreliable prompting to engineered reliability for critical tasks.

Simplified Development - Build complex agents faster by relying on guaranteed structural components and predictable execution flow.

Increased Trust & Control - Confidently delegate tasks to AI through observable, controllable, and reliable performance.

Built-in Agent Capabilities

Internal Reasoning

A dedicated state for planning and reflection, separate from external communication.

Stateful Scratchpad

Persistent space within the agent's state for intermediate calculations or notes during execution.

Guaranteed Tool Execution

Reliable interaction with external tools and APIs, with parameter validation enforced by the Proxy Structuring Engine.

Adaptive Capabilities (MCP)

Dynamically load new tools at runtime via the Model Context Protocol without interrupting the agent.

Getting Started

Install the `proxy-base-agent` library from PyPI. Launch the interactive setup wizard to configure and run your first reliable agent.

pip install proxy-base-agent

Launch your first agent via the interactive setup wizard.

python -m agent