Self-calibration sandbox · Public beta

Thoughts that think back.

An instrument for watching your own thinking — a dialectic between two voices, your library reflected back at you, and a sandbox of small widgets you build, fork, and share. Sovereignty over your own pattern, not advice.

Should I take the Berlin job?
FForm · grounded
low-temp

MMotion · sign-inverted
high-temp

Synthesis · geometric mean
T
E
R
A
harmonizing breath ratios…

Mock dialectic · two passes, one synthesis · loops every ~10s

THE INSTRUMENT

Three pillars. One sandbox.

Thoughts is not a chatbot, not a coach, not a journal app — it is an instrument for self-calibration. The three pillars are what make the difference visible.

Twin-inference dialectic

F-pass + M-pass, one synthesis

Every prompt runs twice — once grounded (form, low-temp) and once sign-inverted (motion, high-temp). The geometric-mean merge becomes the answer. Hallucination becomes legible: when the two passes diverge, the divergence is the signal. You can promote either pass, regenerate one, or remix the synthesis.

Customware marketplace

Personal calibration tools, shared

Every widget you build lives in your own L2 with full git history. Promote one to the members library and the original byline travels with it. Reviewers calibrate; authors keep credit. Versioning is real git tags — pin to a commit, see diffs, fork freely. A network of instruments, not an app store.

Your library, mirrored

Reflection grounded in your own corpus

Write a journal entry; a side panel reflects it back through the books you've actually uploaded. Real page citations, click-through to the page, every reflection persisted as part of the entry's history. The reflection is your library reflecting at you — never a generic LLM opinion.

SHIPS DAY ONE

Three starter skills.

Working examples of what the platform is for. Use them as-is, or fork the source and bend them to your own pattern.

Oracle reading widget

256-card spread · durable artifact

Type "do me a 3-card spread on whether to take the Berlin job." The agent draws from the 256-card archetype set, generates an interpretation, and persists the reading as a one-off widget in your L2. Six months later, you can come back to the Berlin reading, see exactly what was said, fork it, or add a follow-up.

Library-grounded journal

Your books, reflecting at you

Write the entry. On save, a side panel auto-generates a reflection that cites pages from the books you've uploaded — real citations, click-through to the source. The reflection is persisted as part of the entry's history, so the conversation between you and your library accumulates over time.

Listening-party room

Spin a room, fork the template

Say "spin up a room for the new drop, invite-only, with the resonance visualizer." The agent generates a one-off room widget with a dedicated URL, custom visualizer config, and guest list. Fork the template into your L2 to customize, then submit it to the marketplace.

HOW IT RUNS

Free in your browser. Deeper in the cloud.

The free tier is honest about what it is — your local model on your device, fast and private. The Companion tier opens the full twin-inference dialectic across two distinct models, where the divergence has somewhere to go.

Free · Seeker

Run it on your hardware

In-browser local model · zero egress

  • Your local model runs entirely in the browser tab
  • Both F-pass and M-pass on the same local model — divergence is naturally smaller
  • Full access to the customware sandbox, your L2, and starter skills
  • Library RAG and journaling against your own uploaded corpus
  • Model name and size shown at runtime; nothing pre-downloaded without your click
Start free
Companion · Cloud-routed

Open the full dialectic

F and M routed to distinct cloud models · widest divergence

  • F-pass and M-pass routed to two different cloud models for real dialectic spread
  • Higher-fidelity Oracle interpretations and library reflections
  • Listening-party rooms and shared marketplace promotion
  • Author attribution stipend eligibility for promoted widgets ($5/mo at 100 MAU)
  • Switch to local-only mode any time — server-validated, no silent downgrade
See Companion
Thoughts is a self-calibration sandbox, not a clinical or therapeutic product. How we hold the framing.
THE THREE-FACE SUITE

Three surfaces. One reasoning substrate.

Thoughts is the lens you reflect through. AgentStudio is where you author the instruments. AnythingLLM-AAMT is the corpus the other two read from. Use one. Use all three.

Sovereignty over your own pattern.

Your readings are yours. Your journal entries are yours. The widgets you build are yours, with full git history and forkable source. The dialectic is the instrument; what you do with it is the practice.