Stop Predicting and Start Reasoning¶

LLMs Pretend to Think¶

LLMs often struggle with complex instructions and may hallucinate at unexpected times. As predictive language tools, they are unable to truly exercise memory, logic, or arithmetic. This poses significant risk when using LLMs for decision making, especially as the reason behind an LLM's response is often unclear.

With larger prompts, context, and complexity, the shortcomings of LLMs become more apparent.

LLM Reasoning Failure

Prompt:¶

As a bakery, I want to buy 10 apples if I have less than 10 in inventory, but only if my supplier has apples used for baking in stock. Given I have 9 apples, and my supplier has "Honeycrisp", how many apples should I order?

Failed AI Response:¶

Okay, so if you have 9 apples, how many apples do I have? Hmmm... If I have 2 apples and I give them to you, then you will now have 12 apples. So in that case, you should order 0 apples.

Use Logic, not Guessing¶

Vulcan solves these problems by combining the best of LLMs with the best of rule-based decision engines. By separating logic, retrieval, and LLM prediction into composable units, Vulcan allows precise control over complex reasoning and decision tasks for more predictable and reliable AI results.

Unlike with traditional prompting, Vulcan ensures that important facts are not forgotten and logical operations are always repeatable and explainable.

Read The Docs

Vulcan Hybrid Reasoning

# Use natural language for prediction and data retrieval:
engine.rule(
    when=condition(f"Are {Apple.kind} considered good for baking?"),
    then=action(Apple(baking=True)),
)

# Use computed logic for operations that must be correct:
engine.rule(
    when=condition(lambda: Apple.baking and Inventory.apples < 10),
    then=action(Order(apples=10)),
)

# Intelligent on-demand rule evaluation:
engine.fact(Inventory(apples=9))
engine.fact(Apple(kind="Honeycrisp"))

print(engine.facts)
> {'Inventory': Inventory(apples=9),
    'Apple': Apple(kind='Honeycrisp', baking=True),
    'QueuedOrder': Order(apples=10)}

A robot confused by choices.

More Reliable Than Agents¶

Agent-based approaches rely on LLMs to correctly decide when to call external tools. As a result, LLMs may fail to call the correct tool with the correct parameters, and fail to understand the results.

Vulcan separates the processing of strict logic conditions from AI inferencing, improving overall performance and reliability.

A robot blocked from hacking into a computer.

More Secure Than Guardrails¶

Popular guardrail frameworks constrain an LLM's input and output to harden and control behavior. But due to the nature of LLMs, filters and careful prompting can never prevent all unexpected or dangerous LLM behavior.

Because Vulcan uses computed logic with a microprompting approach, the attack surface is greatly reduced and AI decisions are auditable.

An infographic depicting small prompts.

Simplify with Microprompting¶

Vulcan's design favors microprompting rather than lengthy, complex prompts. Microprompting allows for fine-grained and automatic optimization of retrieval and inferencing tasks, reducing overall token count and enabling parallelization.

Fewer tokens mean lower costs and faster processing for your application.

A robot explaining itself on a chalkboard.

Superior AI Explainability¶

LLMs are inherently black boxes executing a predictive approach to logic. This makes it difficult to explain the reasoning and data behind a decision.

Vulcan's rule-based computed logic and variable-based microprompting provides a clear explanation for every decision.

A mosaic of 3rd party LLM logos.

Open Source, Open Ecosystem¶

Vulcan integrates with LangChain to level-up virtually any LLM application with improved performance, reduced costs, complex reasoning, and more.

Best of all, Vulcan Core is free for all to use and modify.

Read the Model Config Guide

Much More to Discover:¶

Focus Your LLM

LLMs are powerful predictive tools, but are prone to hallucinations. Vulcan can help your LLM application become more repeatable and consistent.
Reduce Costs

Why evaluate instructions and context that are irrelevant to the task? Dynamic prompting with Vulcan can reduce overall token count and LLM usage costs.
Express Complex Tasks

Complex multi-step reasoning is difficult for LLMs. Vulcan combines the best of LLM with the best of stepwise imperative instruction.
True Traceability

Meet compliance and customer requirements for traceability and explainability. With every decision point discretely defined, determining why an AI made a decision is easy.
Easy to Use

Built in Python, Vulcan is designed to be easy to use and integrate into your existing AI workflows.
Open Source

Let's build the future of AI together! Vulcan Core is open source and free to use.

View the Code

AI-Hybrid Rules Engine for Logical Reasoning.