Mikoshi Curiosity

Find What You Didn't Know You Were Looking For

A domain-agnostic exploration engine. Not recommendations — discoveries. Plug in any state space and find the interesting things.

🔭 Try Curiosity Live pip install mikoshi-curiosity GitHub →

How It Works

Define → Explore → Discover

Step 1 — Define

Define Your Space

Datasets, text, graphs, APIs, games, markets — anything that can be a state space.

Step 2 — Explore

Curiosity Explores

Intrinsic motivation, Go-Explore memory, and prediction errors drive autonomous exploration.

Step 3 — Discover

Get Discoveries

Results ranked by novelty, surprise, diversity, and serendipity.

Exploration Strategies

5 ways to find the interesting

🔍

Novelty

How different from everything seen before. Maximises coverage of unexplored territory.

⚡

Surprise

How much it violates predictions. Finds things that break your mental model.

🌐

Diversity

How different from current discoveries. Ensures breadth across the space.

✨

Serendipity

Unexpected AND relevant. The happy accidents that change everything.

⚖️

Balanced

Weighted combination of all strategies. Best general-purpose exploration.

Built-in Contexts

Explore anything

📊

Dataset Explorer

CSV/DataFrame exploration. Finds anomalies, clusters, and unexpected correlations.

📝

Text Explorer

Documents, papers, corpora. Discovers bridging concepts between fields.

🕸️

Graph Explorer

Networks of any kind. Finds bridge nodes, structural holes, hidden communities.

🔢

Numeric Explorer

Parameter spaces. Discovers phase transitions, sweet spots, critical thresholds.

🔌

API Explorer

Steam, Spotify, arxiv, any external source. Explores APIs autonomously.

🛠️

Custom

Implement the StateSpace interface for anything. If it has states, Curiosity can explore it.

Quick Start

Five lines to discovery

from mikoshi_curiosity import CuriosityEngine
from mikoshi_curiosity.contexts import DatasetSpace

space = DatasetSpace("sales_data.csv")
engine = CuriosityEngine(space, strategy="balanced")
result = engine.explore(budget=200)

for discovery in result.top(5):
    print(f"{discovery.reason}: {discovery.state.metadata}")

Comparison

Why Curiosity wins

	Curiosity Engine	Recommendation Systems	Random Search
Finds unexpected things	✅ Yes	❌ Confirms bias	Sometimes
Learns from exploration	✅ Yes	✅ Yes	❌ No
Domain-agnostic	✅ Yes	❌ Domain-specific	✅ Yes
Remembers interesting states	✅ Yes	❌ No	❌ No
Explains discoveries	✅ Yes	Limited	❌ No

215

Tests Passing

Strategies

Built-in Contexts

Apache 2.0

License

Open Source Engine + AI = Iterative Discovery

The mikoshi-curiosity engine is open source — it scores states for novelty, surprise, diversity, and serendipity. When combined with AI, it becomes something greater: an iterative exploration loop that discovers ideas no single prompt would produce.

How It Works

💬

1. Your Prompt

"Suggest innovations for my fintech platform"

🤖

2. AI Generates

Mikoshi AI (free), Claude, or OpenAI generates diverse initial ideas

🧠

3. Curiosity Scores

Engine scores each idea for novelty, surprise, diversity, serendipity

🔄

4. Go Deeper

Most surprising ideas feed back to AI — "explore THIS direction further"

          Prompt → AI generates → Curiosity scores → Most surprising → AI explores deeper → Curiosity scores → ...
        

🆓

Mikoshi AI (Free)

🟣

Claude (Anthropic)

Best reasoning. Deeper, more creative exploration with stronger connections between ideas.

🟢

OpenAI (GPT)

Fast generation. Great for high-volume exploration across wide state spaces.

Why Multiple Rounds Matter

Round 1 — The Obvious

AI gives you sensible suggestions. Dashboard improvements, mobile app, notifications. Stuff you'd get from any prompt.

Round 2 — The Surprising

Curiosity finds the weird outliers from Round 1. AI explores THOSE directions. "AR City Planning Playground"? Tell me more.

Round 3+ — The Discoveries

Ideas that no single prompt would produce. Combinations, extensions, and insights from deep in the exploration space.

Nexus Feature

AI-Powered Exploration requires Nexus

The open source engine is free forever. The AI integration — iterative Go-Explore with Claude, OpenAI, or Mikoshi AI — is a Nexus premium feature.

Nexus gives you access to Curiosity AI, Synapse multi-agent, Turbo verification, and the full Mikoshi platform.

£24.99/month

All Nexus features included

Upgrade to Nexus → Try Demo (Free)

The exploration engine is open source. Build your own integrations.

pip install mikoshi-curiosity View on GitHub →

215 tests · 5 strategies · 5 contexts · Apache 2.0