Automates triage, planning and delivery tracking, and runs on our own work every day.
- Automatic task triage and classification
- Delivery tracking across projects
- In daily use on our own delivery
- Open source, under the linear-agents repo
All your answers are in here. Somewhere.
JOINTHUBS
The joint finds what the hubs hold.
JOINTHUBS
The joint finds what the hubs hold.
Delivered, not proposed.
Nothing here is a mock-up. Open a card to see how far it got, then let’s build the next one together.
PUBLISHED
Open source and peer-reviewed. Open any of it right now.
Your notes stay on your own machine, and ThoughtMap serves them over MCP from Docker, so a coding agent reads and writes the knowledge base directly instead of guessing at it. Any harness can connect, Claude Code included.
- Local-first: the vault never has to leave your machine
- ThoughtMap serves it over MCP, in Docker
- Open to any agent harness, Claude Code included
- Semantic clustering, local Ollama embeddings
A library for building features that carry their own context, with validation patterns that do not leak.
- Context-aware feature engineering
- Leakage-safe validation patterns
- Hierarchical statistical signals
- Published on PyPI
A methodology for measuring whether a prediction actually earned its keep, and peer review to back it.
- Return on Expectations metric
- Timeseries validation protocols
- Performance benchmarking
- Published in Risk.net
PROVEN AND PORTABLE
The research is done and the parts already exist. Bring a problem shaped like one of these and the same method goes straight in.
Models how ownership is split, then reports on pay in the format the transparency rules ask for.
- Equity and cap table split modelling
- Pay transparency reporting for HR
- Audit-ready report generation
- Evaluated by two Katowice law firms
One aggregated model, nested by category, with a separately computed model for every district rather than a single city-wide average.
- One aggregated data model, nested by category
- Separate computed model per district
- City down to district granularity
- Strong validation results
Reusable integration blocks that connect models to the tools a team is already working in, without a rebuild.
- n8n-based workflow automation
- Wires AI into systems already in place
- Reusable integration building blocks
- No lock-in to a single model provider
READY TO RUN
Working products, waiting for a system to join. Deployed into your stack, or carried in on a drive and running the same afternoon.
Interpreting EEG recordings with AI, so the slow manual part of the read happens automatically. The MVP has been through its first round of testing and ships today as an educational tool: clinical use waits on registration as a medical device, which is a long road.
- Automated interpretation of EEG recordings
- MVP through its first round of testing
- Available now as an educational tool
- Clinical use pending medical device registration
A workflow intelligence platform packaged to run locally, with the security and auth work already done.
- Runs fully local in Docker
- Security, auth and data-in-transit baked in
- Packaged models, n8n hooks, Python scripts
- Optional cloud GPU deployment
Engagement Tiers
From a single conversation to full production.
Discovery
Frame the real question.
€100 / h
Consultation
- Problem definition
- Feasibility assessment
- Initial roadmap
Prototype
Validate the idea on real data.
From €2,000
< 3 months
- Functional MVP
- Custom model tuning
- Integration testing
- Metrics report
Production
Deploy, scale, maintain.
From €5,000
Scoped engagement
- Enterprise SLA
- System integration
- Compliance & security
- Continuous training
BEFORE YOU ASK
The questions you are actually asking.
You work across property, contracts, clinical data and delivery. Does that not make you a generalist?
The domains differ. The method does not. Each one runs the same steps in the same order, and the domain knowledge comes from your team. What I bring is the part that is identical every time, which is exactly the part most projects get wrong.
Who owns the code when we stop working together?
You do, from day one. Your cloud project, your database, your repository. Nothing runs on my login.
You have no client testimonials.
Correct. I have systems you can open and use instead. Section one is the list.
Our data cannot leave the company.
Then it does not. Local-first is the default, not the upsell. Neo runs entirely in Docker on your own machine, and Jointhubs OS embeds locally with cloud calls only when you switch them on.
What does it cost to find out if this is worth doing?
One hour at €100. If the answer after that hour is that you do not need me, you have paid €100 to avoid a €20,000 mistake.
Tell me what is broken.
One message. I will tell you honestly whether this is a real problem for AI, a plain software problem, or something you should not build at all.