Current experiments
Tools I build for myself and ideas I’m still working through
Personal apps
Tools I build for things I want to keep track of, understand or come back to.
Budget
Bank transactions, investments and manually tracked assets in one place. I use it to look at spending and net worth without piecing everything together each time.
Imports and categorisation handle the everyday work. Transfers, reimbursements and missing exchange rates need a bit more care.
Radar
A research tool for spotting product opportunities. It collects market signals, keeps the sources and records why an idea might be worth a closer look.
Research runs have budgets and checkpoints. I can return to a question with new evidence without starting from scratch.
Personal memory
Voice notes, searchable knowledge and location history. A way to get a thought down, find it later or look back at a trip.
Voice transcription and embeddings run locally. Financial records and precise location history stay private.
Infrastructure
The apps run on my home server. OpenClaw is the assistant I use day to day; the rest of the setup gives the work somewhere to run and somewhere to live.
- Use
- Budget · Radar · OpenClawPersonal apps and the assistant interface
- Connect
- APIs · MCP · Background jobsDefined operations for assistants; imports and research that keep running after I close a browser tab
- Keep
- Forgejo · Anytype · App dataCode and review history, project knowledge and records that outlive a chat
Private networking, routing, monitoring, logs, backups and keeping public services separate from personal ones are part of running it.
I also experiment with local model inference; most reasoning and coding still uses hosted models.
Software Factory
I’m testing agents for research and software delivery I can review.
Research
Ask a specific question. Keep the sources.
Build
Give an agent a clearly scoped task.
Check
Test the change. Keep results with the code.
Review
A person approves publishing, outreach and spending.
Making agent work survive the chat
I keep code revisions, check results and decisions together, then read stored artifacts back to verify delivery. Checkpoints tell the next session where to resume; my acceptance stays separate from a finished run. I use this for my own software and research. It’s still an experiment, and I fix the workflows when they break.