Preserve
Your words are kept first, as their own source, before AI adds anything.
The origin layer for human thinking
Capture the human
before the model speaks
The model forgets. Origo remembers.
Keep your own words as a source. Explore what AI contributes, then add your response without losing the distinction. Origo is being built so the interaction can be followed back to its sources: what you expressed, what the model added and what you chose. Better thinking together, with the human still in charge.
Origo is in private alpha. The complete thinking sequence is in development.
Pocket Chronicle · iPhone · Targeting November 2026
Origo principles
Connect
Return to the books, experiences, and ideas behind a thought.
Develop
Explore with AI while keeping its contribution visible.
Decide
Keep your response, your change of mind, and your own decision.
Content Credentials, built on the open C2PA standard, attach a signed history to a file: which tool made it and how it was edited. That answers an important question. What touched this asset?
It does not answer a different one. What did this person actually express before anything was generated? Once a person and a model have worked together, the finished text rarely shows where one ended and the other began.
Origo is built for that second question. Your note or voice transcript is saved as its own source, and saving it does not depend on any model. AI answers are separate records that point back to the exact version of your words they used. The complete sequence is being built to keep your responses and decisions attributable to you; that full loop is not implemented yet.
File-level credentials compared with the origin layer
| File-level credentials (C2PA and similar) | The origin layer (Origo) | |
|---|---|---|
| The question it answers | What touched this asset? | What did this person express before AI answered? |
| What it describes | A finished file: an image, video, audio clip, or document, with its edit history. | A person's own words and how they developed: notes, transcripts, responses, decisions. |
| Where AI appears | As a tool that generated or edited the file. | As separate, labeled output, linked to the exact version of the words it used. |
| When it is recorded | When the file is created or edited. | Before the model speaks. Saving your words does not depend on any model. |
The two are complementary. Origo does not replace Content Credentials and does not claim conformance with the C2PA specification.
Who it is for
Use AI openly. Keep your authorship clear.
Anyone who thinks with AI can lose track of what was theirs. For some people, being able to show it matters.
Students and writers
Show that your thinking came first.
Keep your reading notes, ideas, and outlines as your own dated sources, with any AI help labeled as AI's. When someone asks how you got there, you have more than your memory.
Researchers, journalists, consultants, freelancers
Keep your judgment on the record.
Use AI for research and drafting while your own analysis, sources, and decisions stay distinct, so you can show clients and editors what was yours.
Teams building with AI
Add an origin layer to your product.
Capture what your users expressed before your model responds, and keep every generation attributable. Pilot conversations are open.
AI can help you connect ideas, consider alternatives, and see something you missed. Keep those benefits, along with a clear account of what you brought to the conversation.
Chronicle is Origo's record of your expressed thinking and its later development. Your words, dates, chosen context, and corrections remain distinct from machine interpretation.
Over time, the aim is a personal corpus you govern: books, preferences, experiences, and decisions that AI can work with. A model's guess about you should remain a guess you can inspect and contest.
01
Remember why a book mattered.
Keep your reaction beside a confirmed book reference. The catalog description stays attributed to its provider; your response stays yours.
02
Let your preferences evolve.
A favorite, a new experience, a change of mind. Build a record from what you actually express, with room for uncertainty and revision.
03
Return to the reason you decided.
Origo is being built to keep the starting thought, AI suggestions, your response, and the decision together, so you can pick up the thinking later.
The experience we are building
Keep the path from thought to decision.
Human source → AI contribution → human response → further AI contribution → human decision. Each step should remain readable, attributable, and linked to what came before.
- 01
Your thought
Your expression, with its recorded date and capture context.
- 02
AI contributes
Its output, provider, model, time, and exact inputs.
- 03
You respond
Agree, reject, qualify, or adopt. The AI origin stays visible.
- 04
AI continues
The same or another model works with the linked history.
- 05
You decide
Your judgment becomes a new contribution, with its connections kept.
The goal: Origo remembers the thinking after the model forgets. Reopen the saved sequence, inspect its inputs, and continue in a new session. The complete loop, provider attribution, and resumption across models are development goals, not shipped capabilities.
See what is implemented todayA concrete example
A better decision. And a way back to why.
You're considering a new role. A book about time has made you rethink what a good week looks like. AI helps you explore the tradeoffs. You keep the parts you agree with, challenge the rest, and make your own decision.
Months later, the useful record is the path you took: your original priorities, the suggestions, your reservations, and what you chose.
Illustrative scenario with invented words and dates. Shows the intended Origo experience, not a product screenshot or a completed feature.
One decision, five contributionsPlanned experience
- Your wordsDay 1 · 8:10 am
The starting thought
I want work that stretches me, but I don’t want to give up every evening. That book reminded me how much unhurried time matters.
Human source · Kept text and chosen book context
- AI contributionDay 1 · 8:12 am
A possibility to consider
You could compare the roles by learning opportunities and the time they leave outside work, rather than by title alone.
Example model A · Linked to your request and starting thought
- Your responseDay 1 · 8:15 am
Keep the idea. Add a condition.
I’ll use those two criteria. But I need to ask about the actual workload; a flexible-work promise isn’t enough.
Human adoption and qualification · AI origin retained
- Further AI contributionDay 2 · 6:30 pm
Continue with another model
Ask the team how often they worked late last month, and what a normal week looks like. That tests the concern you recorded.
Example model B · Linked to the prior thought, output, and response
- Your decisionDay 2 · 6:35 pm
The next step is yours
I’ll talk to two people on the team before deciding. More interesting work matters to me; having evenings still matters too.
Human decision · Linked to the contributions it responds to
Inspect the planned provenance
- Human source
- Original expression, recorded capture and keep times, capture method, chosen context, and later versions.
- Machine output
- Provider and model identity, generation time, request and prompt version, and the exact source and prior output versions supplied.
- Human response
- The contribution you explicitly keep and what it responds to. Adoption does not erase the machine origin of an idea or wording.
- Continuity
- Preserved contributions and explicit links, intended to reconstruct relevant context after closing a session or changing models.
These are requirements for the complete Origo sequence. This illustration does not demonstrate a working provenance inspector or disclose a model's hidden internal reasoning.
What is implemented today
Private alpha · Not generally available
A working foundation. A bigger loop ahead.
The web alpha has source capture, retrieval, routed AI generation, saved answers, and human continuation links. Its Ask path does not yet resume the full linked human and AI history. Complete provider attribution and the durable five-step sequence remain work to be done.
- Text and approved voice-transcript capture
- Human sources kept separately from machine interpretations
- Ask answers stored with a model identifier and links to retrieved source versions
- Human continuation links and source corrections that preserve earlier versions
- Chronicle root witnessed outside Origo, and checked from an operator export Verification material
Pocket Chronicle · For iPhone
In development · Targeting November 2026
Your thinking, close at hand.
A personal Chronicle that keeps your record encrypted on your iPhone. Capture a thought, add context if you want, and return to the words you kept. No account, backend, or cloud synchronization.
Optional Apple labels run on device where supported. Book lookup is an explicit exception: a submitted query and selected book ID go to Open Library; your thought text does not.
Ask about the November releaseRelease target, subject to validation. Pocket Chronicle has not launched.
Implemented in development builds
- Typed thoughts, encrypted unfinished drafts, and explicit Keep
- On-device voice transcription for review before Keep, where supported
- Encrypted local storage, thought history, dates, and declared context
- Local search across words, labels, context, and confirmed book references
- Human-declared links, possible connections, and source-linked review
- Optional rules labels and opt-in Apple on-device label comparison
- Password-protected source export and local deletion
In progress and planned
A bounded experiment to ask questions over selected thoughts using Apple's on-device model. Evidence retrieval is implemented; the Ask interface, generated answers, and real-device evaluation are still in progress.
This experiment is designed for session-only answers. It does not deliver the durable Origo thinking sequence. Lock Screen, Control Center, and Share Sheet capture are also planned.
Know what is kept
Current Pocket voice capture keeps a reviewed transcript, not the original recording. Original-audio preservation is future work. Source export currently excludes local context, book attachments, and derived material; import and restore are not available. Keep exported copies safe: deleting on the phone does not delete those copies.
Via Lumina Reader · For iPhone
Preparing for App Store release
A quieter way to read Scripture.
Open a passage, follow a word, or compare translations. Via Lumina Reader keeps Scripture close, with offline reading and a library of passages and searches you choose to save. No account required.
A free reader with its own local reading library on your iPhone. Reader and Pocket Chronicle are separate apps; they do not share an account or automatically exchange your records.
Reader supportApp Store availability will be announced here after release. Privacy policy
In the current release build
- Read the Berean Standard Bible, World English Bible, and King James Version offline
- Find a passage by reference or search words and phrases with book filters
- Compare translations in a parallel reading view
- Save passages, highlight verses, and return to saved searches
- Choose your reading font, text size, and light or dark theme
- Set an optional daily reminder scheduled on your iPhone
- Export and import saved data in a readable, portable file
Building a research tool, a writing product, or an AI-assisted decision workflow? A saved answer is more useful when a reviewer can inspect its inputs, identify the machine contribution, and see what a person actually accepted or changed.
Deep Well Labs is developing the source and provenance foundations behind Origo. We welcome conversations about preserving human authorship, inspecting derivations, and carrying context across model changes in your product.
Pilots are scoped design partnerships. A production Chronicle integration offering, commercial terms, and service guarantees are not available today.
Inspectable evidence
Trace the record. Understand the limits.
Chronicle publishes evidence formats, publisher keys, and verifier build hashes. The web alpha's operator export has been checked offline with external timestamp evidence. Pocket's local export is a different profile and does not carry that external witness evidence.
Integrity evidence can detect changes to a covered record relative to a witnessed commitment. A recorded capture date is not independent proof of when a thought occurred. Neither integrity nor attribution proves a speaker's identity, originality, truth, or legal admissibility.
Inspect the verification materialNew directions from Deep Well Labs
Govern AI Use
A shared need, not a single app or connected service. Watchtower explores observation; Remis Office explores explicit authority for delegated work; Aether explores household boundaries. These companion directions share a commitment to human control, not a common personal-data store.
Watchtower · macOS · In development
Meet the Presence Sensor.
Watchtower is an observe-only window onto the AI services your Mac talks to and the tool calls your coding agents make. It blocks nothing and sends nothing off your Mac. Its job is one question most people cannot answer today: did any of that happen while nobody was asking?
Building now
Presence Sensor
Knows when you’re away. Notices what kept working.
Lock the screen, close the lid, go make coffee. If an agent makes a tool call or an app reaches an AI service while nobody is at the keyboard, Watchtower marks the moment and tells you. It records only that you stepped away and came back. Never what you typed, never what was on screen.
First Contact
The first time an app talks to AI, you hear about it.
Not the hundredth call from the coding agent you chose. The first call from a notes app, a browser helper or a tool you forgot you installed. Watchtower learns what is normal from your own Mac for a week, then speaks up only when something new reaches out.
Standing Grants
Who holds the keys, and when that changes.
Login items, launch agents, browser extensions, agent configurations, files shaped like API keys. Watchtower lists what on your Mac holds a standing permission to see or act, and tells you when the list changes. It never changes a grant, and never asks for one it does not already hold.
Humans and AI are better together. That takes trust, and trust needs a way to check. Watchtower is being built so you can.
Watchtower observes only. It does not block, intercept or promise complete coverage, and it is not a public release. Presence Sensor, First Contact and Standing Grants are in development; each ships only as far as a Mac can observe without new privacy permissions.
Aether · Proposed concept
Choose the boundaries.
Household AI boundaries for people who want calm, understandable controls. Quiet Hours, family settings and phone/home coverage remain concepts to validate, not a protection service available today.
These are distinct Deep Well Labs product directions. They do not monitor your Origo record or automatically share data with Pocket or Via Lumina.
Explore how these directions differ