ReasonKit

The Reasoning Engine — Auditable Reasoning for Production AI

ReasonKit transforms ad-hoc LLM prompting into structured, auditable reasoning chains. This meta-crate provides a unified installation for the complete ReasonKit suite.

One-Line Install

# Install complete ReasonKit suite
cargo install reasonkit

# Or use the universal installer
curl -fsSL https://get.reasonkit.sh | bash

Installing provides two equivalent binaries:

• reasonkit — Full command name
• rk — Short alias for faster typing

What's Included

Quick Start

CLI Usage

Use reasonkit or the short alias rk — they're identical:

# Run structured reasoning (ThinkTools)
rk think --profile balanced "Should we migrate to microservices?"

# Quick 2-step analysis
rk think --profile quick "Is this email phishing?"

# Maximum rigor (paranoid mode)
rk think --profile paranoid "Validate this cryptographic implementation"

# Verify claims with triangulation
rk verify "GPT-4 has 1.76 trillion parameters"

# Start MCP server for AI agent integration
rk serve

Library Usage

use reasonkit::prelude::*;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    // Create reasoning executor
    let executor = reasonkit::core::thinktool::ProtocolExecutor::new()?;

    // Run GigaThink for multi-perspective analysis
    let result = executor.execute(
        "gigathink",
        reasonkit::core::thinktool::ProtocolInput::query("What factors drive startup success?")
    ).await?;

    println!("Confidence: {:.1}%", result.confidence * 100.0);
    for step in &result.steps {
        println!("- {}", step.as_text().unwrap_or_default());
    }

    Ok(())
}

ThinkTools

Five core reasoning protocols:

Profiles

Pre-configured protocol chains:

Features

[dependencies]
# Full suite (default)
reasonkit = "0.1"

# Core reasoning only
reasonkit = { version = "0.1", default-features = false, features = ["core"] }

# Memory layer only
reasonkit = { version = "0.1", default-features = false, features = ["mem"] }

# Web automation only
reasonkit = { version = "0.1", default-features = false, features = ["web"] }

LLM Providers

18+ providers supported out of the box:

• Major Cloud: Anthropic, OpenAI, Google Gemini, Vertex AI, Azure OpenAI, AWS Bedrock
• Specialized: xAI (Grok), Groq, Mistral, DeepSeek, Cohere, Perplexity
• Aggregation: OpenRouter (300+ models)

# Set your API key
export ANTHROPIC_API_KEY="sk-ant-..."

# Or use a different provider
rk think --provider openai --model gpt-4o "Your question"

Architecture

reasonkit (meta-crate)
│
├── reasonkit-core    ─── The Reasoning Engine
│   ├── ThinkTools
│   ├── Protocol Executor
│   ├── LLM Client
│   └── MCP Server
│
├── reasonkit-mem     ─── Memory Infrastructure
│   ├── Vector Storage (Qdrant)
│   ├── Sparse Index (Tantivy)
│   ├── Hybrid Retrieval
│   └── RAPTOR Trees
│
└── reasonkit-web     ─── Web Sensing Layer
    ├── Browser Controller
    ├── Content Extraction
    └── MCP Sidecar

Philosophy

"Designed, Not Dreamed" — Structure beats raw intelligence.

ReasonKit imposes systematic reasoning protocols on LLM outputs, producing more reliable, verifiable, and explainable results.

Documentation

• Website: https://reasonkit.sh
• API Docs: https://docs.rs/reasonkit
• Core Docs: https://docs.rs/reasonkit-core
• Memory Docs: https://docs.rs/reasonkit-mem
• Web Docs: https://docs.rs/reasonkit-web

Individual Crates

If you only need specific functionality:

# Reasoning only
cargo install reasonkit-core

# Memory layer only
cargo install reasonkit-mem

# Web automation only
cargo install reasonkit-web

License

Apache-2.0 — See LICENSE for details.

Contributing

See CONTRIBUTING.md for guidelines.

Turn Prompts into Protocols | https://reasonkit.sh