A Maine company building curated datasets, open-weight language models and practical software. Our work starts with the records people keep, the questions they ask and the tools they need.
Krummholz is the wind-shaped growth of trees at the edge of
their environment.
The forests and mountains that inspire our name also shape the work we choose.
Hayven Jackson, founder, Lead Data Engineer and employee of Krummholz AI LLC, brings experience
at the South Dakota State Data Center (2013–2018), web
development, CRM systems, and AI data workflows to Krummholz.
Explore our datasets and Bristlecone today. Our first open-weight models are available for beta testing, and a broader synthetic data program is in development.
AVAILABLE TO EXPLORE · BRISTLECONE 1.4.2 BETA
See the terrain. Plan the day.
Bristlecone brings trail maps, offline downloads, hike recording, and mountain-weather estimates into one outdoor tool. Built from time on Maine’s trails, with maps covering the US and Canada.
Explore contours, landmarks and land boundaries.
Choose trail types and map layers for your activity.
Switch between light and dark appearances as conditions change.
An early beta, still being tested. Mountain-weather estimates complement official forecasts and your own judgment.
We turn public source collections into focused, documented releases for AI experiments in forestry, fishing, customer service, workplace robotics, and LLM training.
Task-focused data in open, machine-readable formats.
Documented filtering, duplicate handling and privacy screening.
Source attribution, reuse terms and checksums alongside each release.
These releases are preliminary research material. Independent domain review and production validation remain ahead.
Our working dataset catalog. Each release keeps its data and documentation together.
BETA RELEASES · EARLY LAUNCH NOVEMBER 15, 2026 · FULL LAUNCH JANUARY 2, 2027
Models with dynamic weights. Three practical directions.
Release target: less than 1% hallucination on a defined evaluation, alongside quality, privacy and security checks. This target is not yet achieved or a guarantee across all uses.
Our LLMs will be released with open weights under Apache 2.0, with dynamic, user-reviewed learning in development. Run your own model or use paid hosted inference. We will also offer private-data training for individuals and companies. The goal is user independence, not deeper dependence on Big Tech.
The planned “What I learned today” review lets users accept, reject or edit proposed learning. Local models may diverge from hosted versions. Preserving learning through updates remains a research and engineering goal. Explore dynamic weights →
Acadia
Customer-service answers grounded in approved policies, with request routing and human handoffs. Text first; voice and images later.
Atlantic and the smaller Atlantic Lite are available as open-weight betas. Document workflows, permission-checked tools and hosted services remain in development.
Two models are in beta: Atlantic (27B) and Atlantic Lite (4B) for phones. Both moved to v7 on November 9 after their first round of reinforcement learning from AI feedback, and Atlantic now comes in more file sizes for workstations and consumer graphics cards. Acadia and Katahdin will be distilled from Atlantic; our earlier 35B core is deprecated. No hosted service is released yet; each model card states its license, evaluations and limits.
Websites, CRMs and custom tools designed around the way a business actually operates. Our focus is local businesses and the practical work behind forestry, fishing, customer service and applied robotics.
Explore customer intake, case tracking and clearer handoffs, so the next step and its owner are easy to find.
Bring information together.
Plan useful dashboards and structured records for work that currently lives across forms, spreadsheets and separate tools.
Start with the task.
Define the users, workflow and data permissions first, then agree on a small, testable first release and a plan for support.
AI integration
Your business has a way of working. Your AI should know it.
We train and adapt our models around a business's products and procedures, and advise on AI integration, data retention, data lakes, data pipelines and agentic workflows. Projects start with an agreed scope and a written quote.
krummholz
