Capture First
Thoughts, bugs, decisions, and pivots should be saved before they are polished.
Local-first AI workflow kit
Continuum Kit
Continuum Kit is a way of building personal AI tools that start with real life: messy thoughts, phone dictation, half-finished ideas, agent handoffs, private workbenches, and the practical need to recover when a machine, browser, or conversation disappears.
What It Is
The working thesis is simple: a person should not have to keep the whole system in their head. The computer should capture useful traces, organise them into durable context, and make it easy for a human or an AI agent to pick the work back up.
Continuum Kit is the reusable version of that idea. It is being grown from a private, local, heavily dogfooded workflow before being cleaned up for other people.
Philosophy
Private By Default
Raw personal data stays local. Public material is reviewed, redacted, rewritten, or synthetic.
Use Before Abstracting
Build for the real human first. Extract reusable pieces only after they reduce brain load.
Friction Is Evidence
Lost drafts, confusing UI, crashes, and restarts are product signals, not embarrassing noise.
Parts So Far
| Part | What It Does | Why It Matters |
|---|---|---|
| Phone thought loop | Voice, text, reactions, photos, and routes from a phone into active AI work. | Lets the human keep moving without returning to a laptop for every thought. |
| Workflow Manager | A chairman-style agent thread for planning, capture, triage, and priorities. | Keeps chaos from becoming lost context. |
| Agent launcher | Named commands and tmux sessions for starting the right agent in the right folder. | Removes model-level and resume-hash memory burden from the human. |
| Private workbench | Local HTML reports for browsing logs, pivots, research queues, and story candidates. | Turns raw traces into usable thinking surfaces without publishing them. |
| Continuum Core | The deeper local-first memory, source-log, and personal-system layer. | Provides the long-term substrate for continuity across tools and devices. |
| Public story/blog | Clean static pages that explain the philosophy and publish edited evidence. | Makes the work understandable to other people without leaking the private workspace. |
Long-Term Goals
One-click bootstrap
A non-technical person can install a useful local-first setup without learning the whole stack first.
Safe personal context
Private data, secrets, device state, and raw captures stay protected by default.
Visible agents
People can see which helper is working, what it owns, and how to restart or hand off safely.
Better thinking surfaces
Voice, text, reactions, photos, docs, and logs become one low-friction way to shape thought.
Reusable working agreements
Repeated human-AI collaboration patterns become named helpers rather than fragile prompt folklore.
Public learning trail
The blog explains what worked, what broke, and what can be reused by someone else.