∇f(x*) = 0

arXiv:2412.xxxxx

∀ε>0, ∃δ>0

Σᵢ xᵢ = μ

📄 NeurIPS 2025

🔬 Lemma 4.2

P ≠ NP ?

E[X] = μ

📊 Figure 3

∫₀^∞ e⁻ˣ dx = 1

Open Source · Local-First · MIT License

💡

📄

∂L/∂θ

EurekaClaw
The Research Claw 🦞

The AI that catches your Eureka moments. Scrapes arXiv, generates lemmas & theorems, proves ideas, writes LaTeX papers, runs experiments — all from your chat or terminal. Open source. Local-first. Private by default.

Install in 1 command Star on GitHub

eurekaclaw — bash

$ eurekaclaw prove "Prove a generalization bound for sparse transformer attention."

🦞 Building proof plan and retrieving relevant lemmas...

📄 Found supporting assumptions from prior sparse attention bounds.

💡 Proof skeleton generated: decomposition + covering argument + low-rank approximation step.

✨ Main theorem drafted. Proof outline complete.

🦞 Eureka! Proof notes saved to ./proof_sparse_attention.md▊

Quick Start

Zero to Eureka in seconds 🦞

One command. Any platform. No PhD required (but it helps).

eurekaclaw — 💻 macOS / Linux

# One-liner install

$curl -fsSL https://eurekaclaw.ai/install.sh | bash

🐚 EurekaClaw Installer

Multi-agent theoretical research system

✓ Detected OS: macOS / linux

[1/3] Preparing environment

✓ Python found: Python 3.11+

✓ Git found: git version 2.x

[2/3] Installing EurekaClaw

✓ Repository cloned to ~/.eurekaclaw

✓ Virtual environment created: ~/.eurekaclaw/.venv

INFO Installing EurekaClaw...

✓ EurekaClaw installed into virtual environment

✓ Shim installed: ~/.local/bin/eurekaclaw

[3/3] Finalizing

✓ Seed skills installed to ~/.eurekaclaw/skills/

🐚 EurekaClaw installed successfully!

# Start your first proof

$eurekaclaw prove "Prove a generalization bound for sparse transformer attention."

Requires Node.js 18+ or Python 3.11+. Works with Ollama (local), OpenAI, Anthropic, or any OpenAI-compatible API. 🦞

🦞 Eight Claws

What EurekaClaw Does

Every claw built for one purpose: catching your next research breakthrough.

Literature

🔍

arXiv & Literature Crawler

Fetch, summarize, and cross-reference papers from arXiv and Semantic Scholar. Goes beyond abstracts with configurable PDF deep reading to extract reusable results, assumptions, and proof ideas from the most relevant papers.

Discovery

💡

Eureka Idea Generator

Brainstorm research breakthroughs and novel hypotheses by synthesizing patterns across thousands of papers. Catch connections humans miss.

Workflow

🧩

Researcher-Style Proof Workflow

Builds proofs the way researchers do: through planning, skeleton construction, lemma extraction, verification, and final theorem assembly in a staged theorem-proving pipeline.

Formal Math

🔢

Theorem & Lemma Prover

Generate, refine, and verify mathematical proofs with memory-guided planning, reusable prior results, and optional Lean4 checks. From conjecture to structured theorem, proof skeleton, and verified lemmas.

Memory

🧠

Research Memory System

Combines session memory, cross-run persistent memory, theorem knowledge graphs, and reusable proof hints to support literature analysis, proof planning, and theorem development over time.

Experiments

🧪

Experiment Runner & Code Exec

Writes, runs, and evaluates ML experiments. Integrates with Jupyter, Weights & Biases, and your existing codebase. Hypothesis → result in one loop.

Publishing

📄

LaTeX Paper Writer + Figures

Drafts full camera-ready papers with structured sections, theorem environments, auto-generated figures (matplotlib / tikz), and correctly formatted citations.

Privacy

🖥️