Contact Us

EganAI

We build custom AI, and we prove it works.

Model merging, prompt optimization, agent evaluation, and the products that run on them.

Start a project
1,201
GitHub stars
3
Open models published
195
Tasks in our agent benchmark
7
Products in production

What we do

Four ways we work, each with a measured result behind it

Custom app development

We build the product around the model, from a Shopify embedded app to a native iOS release.

Three Shopify apps and two iOS apps shipped. PadelReplay and Keeproom are live on the App Store.

Custom model merging

We merge open models into one that fits your workload, then prove it on your benchmark instead of a public one.

Last run beat its base model, Gemma-4-E4B, by 16.5% and the best donor by 9.3%.

Prompt and workflow optimization

We tune the prompts and the control flow around them, scored on a held-out set so the gain survives contact with real traffic.

Top-3 prompt ensembling improved the official GEPA library on 6 of 6 held-out tasks (sign test, p≈0.016).

Agent evaluation

We work out which models actually hold up inside your agent, on your tasks, in sandboxes that match production.

Our suite runs 195 tasks across three difficulty tiers in Docker.

Models

Merged LLMs we publish on Hugging Face

View All on Hugging Face →
9BText Generation

qwen3.5-9b-terminal-merge

Terminal-optimized merge, evolved against real command-line tasks.

Available on Hugging Face
4BText Generation

Qwen3-4B-Thinking

Merged for chain-of-thought reasoning.

Available on Hugging Face
4BText Generation

Qwen3-4B-Instruct

Merged for instruction following and general tasks.

Available on Hugging Face

Every model here is a merge, not a fine-tune. We evolve the recipe against real terminal tasks, then publish what wins.

Open Source

Tools and research we maintain in public

View All on GitHub →
awesome-cli-coding-agents

awesome-cli-coding-agents

★ 1,020

The directory of terminal-native AI coding agents and the harnesses that orchestrate them.

GraphMemory

★ 156

A GraphRAG database with hybrid graph and vector retrieval in one engine.

LLM-Shortcut

★ 25

Reach any LLM straight from an iOS shortcut, backed by a Docker endpoint.

In production

Products we designed and built end to end, all live today

BreadCode

BreadCode

Self-hosted software factory for coding agents

breadcode.app
Crontender

Crontender

Scheduled agent runs, from a menu bar extra

crontender.com
fittings

fittings

150 utility APIs agents pay for per call, over x402

fittings.sh
Upsella

Upsella

SMS upsells for Shopify, fired after checkout

upsella.com
Virtual Try-On

Virtual Try-On

AI fitting room for Shopify storefronts

aivirtualtryon.shop
Keeproom

Keeproom

On-device AI journal, live on the App Store

keeproom.app
PadelReplay

PadelReplay

Padel match analysis on iPhone

padelreplay.app

Tell us what you need built

Send us the problem and the benchmark you care about. We will tell you whether we can move it, and by how much.

Get in Touch

© 2026 EganAI, LLC. All rights reserved.

PrivacyTerms