Overview
Live from the tm-611 Firestore. live tiles read now; locked tiles need one rule deploy.
users collection (so total users + growth are live), but messages and signups are owner-only. To light up recordings + signups + retention, deploy this one-line-per-collection founder rule (drafted in firestore.rules.founder-analytics.txt):
allow read: if request.auth.token.email in [ "gv@mementote.co","glalonde5@gmail.com","glalonde5@icloud.com"];
Time Machine Catalog
The whole thesis, distilled to ~97 one-line claims — each one restored to the argument behind it. This is the founder's reference: every terse line, with its nuance, its reasoning, and where relevant its evidence or its honest caveat. The science panel is graded exactly as the white paper grades it — established, plausible extrapolation, or product hypothesis. Intellectual honesty is the point.
Claude sees the Catalog text as context. With text selected in Edit mode, it rewrites the selection; otherwise it answers below.
Thesis
™ is not for you. ™ is for space. Every other recording system pulls a memory out of the world and hands it to a person on demand. Time Machine inverts the ownership: the recording belongs to the place, not the user. That single flip is the product — it reframes "a note-taking app" as infrastructure for a medium addressed to locations rather than to people.
Media for the place, not the person. A song is authored for an audience; a ™ is authored for a coordinate. The place is the intended recipient and the gatekeeper of access. This is what makes the category new rather than a feature — it is a different addressee, not a different UI.
Space media: here-and-then. Broadcast media is "anywhere, anytime." Space media is the opposite constraint made into a form: a specific here and a specific then, recoverable only by being in that here. The value comes precisely from what it refuses — ubiquity.
The place is part of the memory. Not a metaphor. A memory carries a where at the level of neural representation (see Science): spatial coding and episodic memory share the same medial-temporal circuitry. Re-presenting a memory at its where works with the grain of the brain rather than against it.
Presence is the access key. There is no password, no share link, no download. Physically being inside the radius is the credential. That makes access embodied and unforgeable — you cannot copy your way into a place you are not standing in.
The Four Verbs
Record. Capture audio on device: voice, music, ambience, or a mix. Capture is deliberately the easy, universal act — the constraint that makes ™ special lives in the next three verbs, not this one.
Leave it in the place. The recording is written at a coordinate with a radius (default 100 m). It is not stored "in your library" — it is stored to a place, which is where it now lives and the only place it can be reached.
Return. To hear it you have to physically come back inside the radius. This manufactures the ritual: a recording becomes a reason to return to a place rather than a file you scroll past.
Play the place. Playback is not "open one file" — it is the place playing back what it holds, mixed and balanced. You do not retrieve a clip; you stand somewhere and the place speaks in the sounds left in it.
The Rule
Play only within the recording's radius. Radius gating is enforced in code: a recording is playable only within radiusMeters of where it was made. The place is part of the memory, so playback requires the place — this is the direct product translation of encoding-specificity (Science §b).
Creator must leave and return to unlock. The "seal": even the person who made it cannot replay it on the spot. They must physically leave the radius and come back. This is what turns capture into a place-anchored, involuntary re-encounter rather than instant rehearsal.
No on-demand access. Withholding on-demand playback is the feature, not a limitation. The vivid, involuntary memory is the one you meet by going back — not the one you request. Convenience would trade that away.
You go to it. It does not come to you. Every other memory system moves the memory to the user. ™ keeps it in the world and makes the user travel. That asymmetry is the source of the scarcity, the ritual, and the reason a place accrues value.
A recording belongs to its place. The core invariant stated plainly. Ownership is spatial. Loop chains never destroy layers; a recording's identity is its coordinate. Break this and you have just another cloud drive.
Leaving It For Someone
For your future self. The default and most reflective case: you leave a recording for the person you will be when you next stand here. The place holds it until you return, unplanned, and meet your own past voice in the setting that made it.
For you and someone. A shared ™ both parties can unlock in place — a joint marker at a shared spot. The place becomes a mutual keepsake rather than a private one.
For someone only. You can leave a recording addressed to another person that you yourself cannot replay — a gift left in the world for them to find by returning to the place.
Given away: the sender loses access. Giving is real transfer, not sharing a copy. When you address a ™ to someone else, you relinquish your own access. Scarcity and the weight of the gesture depend on it being a genuine hand-off.
Addressed by phone number. Identity is a phone number, so you can leave a ™ for someone before they even have an account. (Known gap: a recipient without an account is currently self-addressed silently; the fix is recipientEmail/recipientPending fields claimed at sign-in.)
Understanding
Every recording is tagged on-device. Analysis runs on the phone via a pluggable analyzer system — private, free, instant, no audio leaves the device. This is the on-device-first stance from the ML Report: small models that keep voiceprints and transcripts local.
Tags: key, tempo, mood, scene. Musical and acoustic descriptors so a place knows the character of its sound, not just that "music happened." Today key/tempo is partly heuristic (octave and relative-key errors) — a stated next layer, not a solved problem (see ML Report).
Tags: bird species, speaker, topic. Identity-level tags: BirdNET for species (the Merlin model, on-device, on-brand), voice embeddings for speaker (opt-in, consent-gated), transcription→topic for conversation. The honest gap: we do categories well and identities are the frontier.
A place is queryable by tag. Once every recording is identified along many axes, retrieval stops being "clips near me" and becomes semantic. The place becomes a database of itself that you query by what things are.
"Play the birds here." A worked example: filter a place's audio down to just its birdsong. The place's nature layer, on demand, in place — passive bioacoustic monitoring as a consumer gesture.
"The C-major guitar here." Query by musical feature: every take in a given key, at this spot. This is why key/tempo detection matters — it makes the place sortable by sound, and it is what the generative mix consumes.
"Only the calm ones." Query by mood. The same tagging that files a place lets a returning listener tune what surfaces — calm, upbeat, just voices. The place arranges itself to who is asking.
Play the Space
One continuously generated mix. Standing in radius you don't pick a file — the place produces a single, continuous mix of everything left in it. The output is the place performing its own accumulated history.
Music bed. Voices surface over it. A two-layer structure: ambient beds and music sit underneath, voices and performances surface on top. This is gain-staged by soundType/timbreTags so the mix reads as composed, not piled up.
Tuned by key, tempo, mood. The mix uses the tags to order and blend — matching keys, aligning tempo, grouping by mood — so a place's history plays as a coherent piece rather than clashing fragments.
Recipes: last week, on this date. Temporal filters over the same material: hear only what was left in the last week, or everything ever left on this calendar date. The place as a time-indexed instrument.
Recipes: just birds, just me, upbeat. Semantic filters over the mix: strip to birdsong, to your own recordings, to the upbeat ones. Recipes are the user-facing filing layer on top of the tags.
Shows the mix's key and tempo. The interface surfaces the current mix's musical key and tempo, so a musician can play along with the place — the bridge from listening to performing over it.
Record over the space while it plays. You can capture a new take on top of the live mix, in key and in time — overdubbing onto the place itself. This is where "play the space" becomes an instrument and connects to the loop station.
The Six Layers
1. Space holds sound. The four verbs. A recording written at a coordinate, retrievable only in radius. The base substrate. Shipped — iOS, web, the tmkit SDK.
2. Space requires return. The verbs plus the constraint: leave and come back, no on-demand. This is the layer that makes it media for space, not for you — it manufactures scarcity and ritual. Shipped — the creator seal.
3. Space understands itself. Every recording identified along many axes on-device — key, tempo, species, speaker, topic, mood, scene — and tomorrow the same for video and further modalities. A place becomes a queryable sense-layer. Shipping.
4. A model trains in place only. The identification model lives in the place and only in the place, trained by being kept there on data that only exists there. Tiny, local, and the strongest technical moat. The frontier.
5. Space files and mixes itself. The identifications feed a filing and mixing system — the place is its own archivist and DJ, composing its history into the continuous mix, albums, and tuned retrieval. Shipping.
6. Space composes across senses. The ceiling: outputs integrated across modalities — sound tuned to image tuned to light and beyond, into smell and texture. A place renders a synesthetic experience of itself. The vision.
The Moat
The model lives in the place. Not a general classifier serving every place — a model conditioned on one place as a known prior. Because everything it must explain is "what happens here," it can be smaller and more accurate than any generic model.
Trained by staying there. The model gets better passively, by accretion, from the data that accumulates at that coordinate. A grove learns its own birds; a room learns its own voices; a stage learns its own music. Intelligence grows just by remaining itself.
Only that place's data. Local. Tiny. The training set is the place's own history — few-shot embeddings and priors per location that bias the big models (restrict BirdNET to local species, prime speaker-ID with enrolled voices, prior over key/tempo). Small enough to run on-device.
Uncopyable without the places. A competitor can copy the code but not the accumulated place-priors, because those only exist by having been physically present over time. The moat is not the algorithm; it is the irreproducible local data lodged in real coordinates.
Clients / Substrate
Clients: iOS, web, machines, SDK. One core, many surfaces. The iOS app, the web Time Machine, the physical machines, and the SDK are all clients calling the same four verbs into the same substrate.
The apps are clients, not the product. A common misread is that Time Machine is "an app." The app proves the substrate and touches the world first, but it is one integration among many — the same status a museum guide or a car integration would have.
The product is the substrate. What we sell and defend is the place-media layer for the physical world: store-to-place, retrieve-by-place, tag, and the seal. Everything else is a wrapper over that.
Four verbs any system can call. The verbs stay stable while the wrapper changes per surface. This is what makes it a platform: a stable primitive that anyone's software can call without knowing our internals.
tmkit SDK: record, nearby, play. The developer-facing shape of the four verbs — record a sound to a place, find what's nearby, play in radius. Python/JS today; Swift, C++, Unity next.
Raspberry Pi: a phone left in a place. The hardware reference client — a cheap board running the same verbs, physically installed at a coordinate. It proves the substrate doesn't need a consumer phone; any device that can call the verbs is a machine.
Integrators speak verbs, not schema. Partners call record/nearby/play; they never touch our Firestore schema. That boundary lets the internals evolve without breaking integrations and keeps the substrate licensable as a clean contract.
Where It Plugs In
AR needs place-anchored audio. The big one. Vision Pro and Quest need persistent audio anchored to real coordinates and there is no good substrate for it. The four verbs are that primitive — spatial computing will need exactly this.
Maps: a listen layer. A "listen" tab on a map — stand somewhere and hear what was left there. Maps already own the coordinate; ™ supplies the sound bound to it.
Cars play the places you pass. A car that plays the ™s of the places you drive through — a moving reader of space media. The route becomes a playlist authored by the world, not a person.
Museums: guides left at the exhibit. An audio guide that literally lives at the artwork instead of on a rented headset — left in place, unlocked by standing there. Curators and visitors both leave layers.
Memorials: leave a recording there. A memorial you can leave a voice at, and that others reach only by visiting. The place-gating is exactly right for grief: presence required, nothing on-demand.
Instruments: sample where you are. A DAW plugin or pedal whose source material is the room's own audio — sampling where you are, by key and tempo. Music tooling built on the place as an instrument.
Nature: geo-tagged birdsong. Passive bioacoustic monitoring, geo-anchored by species — the bird ID already ships. A scientific and conservation use of the exact same substrate.
OS: record-to-place as a permission. The endgame: "record-to-place" as a system-level permission, like location or camera. At that point the substrate is infrastructure the whole device speaks, not an app you open.
Physical Machines
Phone booth ™. Hexagonal ™ box. Two signature enclosures. The phone booth elicits addressed, intimate speech; the hexagonal box carries the grid-cell geometry from screen to object. Each form is a prompt — it tells the participant, without instructions, what kind of sound to make.
28-ft geodesic dome, Grizzly Peak. A large-scale install; also the prior-event data channel (deliverable 5) where ingest runs on a real place. Big objects earn attention and seed a place with history before anyone arrives.
Torricelli's Trumpet ™, Kibera. A machine sited in Kibera, Nairobi — proof the substrate is not a Bay-Area toy. Named for the infinite-surface, finite-volume horn: a fitting shape for a place that holds unbounded sound.
Installs at Berkeley parties. Go-to-market as much as art: drop a machine into a live event where people already are, and the recordings accumulate as a by-product of the party. The first event (July 23) is exactly this.
A machine is a phone left in a place. Demystifies the hardware: every machine is the same kiosk app inside a different enclosure, pinned to a coordinate. The form elicits; the device captures; the classifier tags. No per-machine engineering.
Science deepened
Context-dependent memory. The umbrella claim: returning to where something was encoded should help you retrieve it. It rests on two pillars — encoding specificity and environmental context effects — one solid as principle, one real but fragile. Detailed below, with the caveats intact.
Place: space made meaningful. A "place" is not a raw coordinate — it is space that a person has invested with experience. ™ operates on places because that is what memory binds to; the GPS point is just the address of the meaning.
Grid cells: the brain maps space. The literal mechanism under the whole thesis, and the origin of the hexagon. Full treatment below.
"Journals are data." — Dacher Keltner. Keltner's framing (Greater Good Science Center, Berkeley): the traces we leave of our inner life are analyzable evidence, not just sentiment. A ™ is a place-anchored journal entry — data about a self at a moment, made queryable.
Recursive Reflection. The deeper wager: re-encountering your own past sound, fixed to the place that made it, is a reflective loop — the present self meets a past self in the exact setting that produced it. Because the cue is place-anchored and unplanned, it resists the smoothing that flattens a memory into a rehearsed story.
Greater Good fellowship, Berkeley. The reflection/well-being claims are being pursued through a Greater Good Science Center fellowship — the venue for turning the recursive-reflection hypothesis into measured instruments rather than assertion.
The brain does not store what happened separately from where it happened. The same medial-temporal machinery does both — which is the load-bearing fact for the entire product.
- Place cells in the hippocampus fire when an animal occupies a particular location; together they form an internal map of an environment (O'Keefe & Dostrovsky, 1971; O'Keefe & Nadel, 1978). established
- Grid cells in the medial entorhinal cortex fire at multiple locations arranged in a regular hexagonal lattice, tiling space with a periodic metric that feeds the hippocampal map (Hafting, Fyhn, Molden, Moser & Moser, 2005). This work anchored the 2014 Nobel Prize in Physiology or Medicine. established
- The human hippocampus supports episodic memory — memory for personally experienced events set in a spatio-temporal context — using much of the same circuitry it uses for spatial navigation (Burgess, Maguire & O'Keefe, 2002). established
The load-bearing point is the last one: spatial coding and episodic memory are not two neighboring systems — they are entangled in the same structures. A memory has a where built into it at the level of neural representation, so an app that re-presents a memory at its where is working with the grain of the system, not against it.
The hexagonal firing geometry is where the ™ takes its visual and physical form — the hex UI, the hex-pad performance surface, the hexagonal ™ box. We are explicit: the hexagon is a motivated homage to grid geometry, chosen because the same tessellating shape the brain uses to metricize space is a natural unit for a place-based instrument. It is not a claim that a hex button activates grid cells. design framing, not a mechanism claim
Two well-studied ideas say that returning to where something was encoded should help you retrieve it — but they are not equally solid, and we say so.
- Encoding specificity (Tulving & Thomson, 1973): a retrieval cue works to the extent that it matches what was present at encoding. Reinstating the encoding context at retrieval improves recovery. established as a principle
- Environmental context-dependent memory: Godden & Baddeley (1975) had divers learn word lists on land or underwater and recall in the same or the other environment; recall was better in the matching environment. A textbook result — but a 2021 pre-registered replication (Murre) did not reproduce the free-recall advantage. We do not lean on the size of this effect. plausible extrapolation — effect real but fragile
Sound is an unusually potent cue for autobiographical memory. Music-evoked autobiographical memories are typically rated highly vivid, are often involuntary — they arrive unbidden, triggered by an external cue — and carry rich perceptual and emotional detail (Janata, Tomic & Rakowski, 2007; Jakubowski & Ghosh, 2021). In diary studies they occur on the order of once a day. established for music plausible extrapolation to voice, ambience & mixed — the ™ will measure it directly
Put together: an audio cue is a strong involuntary trigger, and returning to the encoding place is a second, independent contextual cue. The ™ delivers both at once — the right sound in the right place. That combination, not either alone, is the design bet.
The deeper wager is about reflection, not just recall. Re-encountering one's own past voice, anchored to the place that made it, is a reflective loop: the present self meets a fixed trace of a past self, in the exact setting that produced it. Because the cue is place-anchored and unpredictable — you meet it only by being there, often when you didn't plan to — it resists the smoothing and rehearsal that flatten a memory into a story. The claim: such unstable, environment-bound auditory re-encounters re-open an episode rather than replay a summary of it, pushing cognition toward adaptivity.
We mark this honestly. The neuroscience in (a)–(b) is not in doubt; the specific claim that place-locked auditory self-encounters produce a measurable reflective or adaptive benefit is the thing the ™ exists to test — pursued via the Greater Good Science Center fellowship, framed as a question, not advertised as a result.
| Scientific point | ™ design decision |
|---|---|
| Memory carries a where; place & episode share circuitry (a) | Radius gating — playable only within radiusMeters of where it was made. |
| Encoding specificity / context reinstatement (b) | No remote playback — retrieval happens in the reinstated context or not at all. |
| Grid-cell hexagonal firing geometry (a) | Hexagonal UI & hex-pad surface — motivated homage carried from screen to physical box. |
| Sound as vivid, involuntary cue (b) | Audio-first capture (voice/music/ambient/mixed) — sound is a uniquely strong trigger. |
| Reflection needs every past voice kept, not overwritten (c) | Layered loop chains — an overdub is a new message with parentMessageId; no layer destroyed. |
| Shared, embodied encounter in real space (a, c) | Machine-embodied recording — a physical object at a pinned coordinate makes place-anchoring literal. |
Where we refuse to overclaim: (1) environmental context effects are real but fragile — Godden & Baddeley is textbook, yet Murre (2021) did not replicate the free-recall advantage, so we treat it as a hypothesis to measure, not a guarantee to advertise; (2) the autobiographical-cue evidence is strongest for music — extending it to voice, ambient, and mixed captures is reasonable but not yet established; (3) the recursive-reflection/adaptivity payoff is a product hypothesis on an established substrate; (4) the hexagon is homage, not mechanism.
What the first event (Del Mar HQ, July 23) can measure — a natural experiment, no lab required: return rate and dwell (cue potency); in-radius vs. attempted remote playback (what the place constraint costs vs. protects); vividness/involuntariness self-report by
soundType (testing the (b) extrapolation); reflection self-report after re-encountering one's own past recording (the first toe-hold under (c)); and layer depth in loop chains when no voice is ever destroyed.
References. Burgess, N., Maguire, E. A., & O'Keefe, J. (2002). The human hippocampus and spatial and episodic memory. Neuron, 35(4), 625–641. · Godden, D. R., & Baddeley, A. D. (1975). Context-dependent memory in two natural environments. British Journal of Psychology, 66(3), 325–331. · Hafting, T., Fyhn, M., Molden, S., Moser, M.-B., & Moser, E. I. (2005). Microstructure of a spatial map in the entorhinal cortex. Nature, 436, 801–806. · Jakubowski, K., & Ghosh, A. (2021). Music-evoked autobiographical memories in everyday life. Psychology of Music, 49(3), 649–666. · Janata, P., Tomic, S. T., & Rakowski, S. K. (2007). Characterization of music-evoked autobiographical memories. Memory, 15(8), 845–860. · Murre, J. M. J. (2021). The Godden and Baddeley (1975) experiment… A replication. Royal Society Open Science, 8(11), 200724. · O'Keefe, J., & Nadel, L. (1978). The Hippocampus as a Cognitive Map. Oxford University Press. · Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80(5), 352–373.
Business
License the substrate. The revenue object is the place-media layer, licensed to integrators — not a consumer subscription for an app. The apps prove it; the license is the business.
Private storage subscription. For individuals: a subscription for private place-anchored storage — your own ™s across the places of your life. The consumer tier that funds and seeds the network.
Public place-gated premieres, ticketed. An artist can premiere work that only exists at a place, for a window — a ticket to a location-and-time, not a stream. Live, place-gated scarcity as an event product.
Split revenue with the artist. Premieres and place-gated releases share revenue with creators, aligning ™ with artists rather than extracting from them. The platform grows as its creators do.
Scarcity rebuilt in physical space. Digital media destroyed scarcity by making everything copyable and on-demand. ™ rebuilds it structurally — a ™ is scarce because it exists in one place and cannot be summoned. Scarcity is the product, re-manufactured.
Per place, per integration, per dev. The licensing meters: charge per place activated, per integration shipped, per developer building on tmkit. Standard platform economics over a novel primitive.
We are selling space and time. The one-line framing of the whole business: not files, not storage — access to a specific here and a specific then. Space and time are the inventory.
Go-to-Market
Piggyback on events and venues. Rather than acquire users cold, drop machines into events and venues where a crowd already exists. The recordings accumulate as a by-product, and the place is seeded with history in one night.
The app that gets you back on it. Retention is structural: a ™ left for you is a standing reason to return — to the place and to the app. The mechanic itself is the growth loop, not a notification strategy bolted on.
Big objects earn attention. A 28-ft dome or a phone booth in a party is its own marketing — physical presence commands curiosity that a download prompt cannot. The machines are acquisition, not just art.
Keep the surprise. Part of the magic is not knowing exactly what you'll meet when you return. GTM and UX both protect that — over-explaining or making playback predictable would spend the very surprise that makes a re-encounter vivid.
Status
Live at mementote.co. The product is public and running — not a deck. The web Time Machine and the brochure are shipped and reachable today.
18 users. 1,005 recordings. Real usage against a real backend. (Total users read live from the users collection on the Overview and Traction tabs; recordings light up with the founder-read rule.)
~56 recordings per user. The signal that matters more than raw counts: high recordings-per-user says the people who are in are genuinely using it, not just signing up. Depth over breadth at this stage.
iOS + web shipping. Two clients live against one core. iOS carries the full record→seal→return→play loop; web carries an iOS-accurate Time Machine with map backdrop, seal, and guest mode.
SDK + Raspberry Pi client built. The substrate story is not vapor: tmkit exists and a Pi client — a phone left in a place — runs the four verbs. Proof the core is separable from the apps.
Provisional patent in prep. The place-gating and place-trained-model ideas are being protected — provisional patent in preparation, handled with G4 on legal.
Team
G: founder, CEO, builds all software. G La Londe V — founder and sole software builder across iOS, web, SDK, and backend. The full stack runs through one person, which is why the roadmap is sequenced the way it is.
Tyler: AI, data, bioacoustics. Owns the model and data side — the tagging, the bird/species work, and the place-tuned-model direction that becomes the moat.
Tyrone: UX research, science. Owns the human side — UX research and the science framing, including how the memory literature maps onto real product decisions and what the first event should measure.
Joe: CTO advisor. G4: legal. Joe advises on technical architecture as CTO advisor; G4 handles legal, including the provisional patent. Advisory depth without headcount.
Cory + Jack: the physical machines. Cory (sound engineer) and Jack (visual engineer + builds) make the machines real — capture pipeline and enclosures for the phone booth, hex box, dome, and event rig.
First Event
July 23. Del Mar HQ. The first event and the deadline that focuses everything — six event machines deployed at a real place, a natural experiment for the memory claims with no lab required.
Record at the door. Leave and return. The event stages the four verbs bodily: capture on arrival, then the seal — you leave the radius and come back to unlock what you and others left. The whole thesis, performed by the guests.
The band plays the room's past. The payoff moment: a live band plays over the in-radius mix — the accumulated audio of the room surfaced as a bed, the place's past performed back into its present. "Play the space," on stage.
Stack
Phone-number identity. Identity is a phone number, which is what makes the Giver mechanic work — you can address a ™ to someone before they have an account. Phone auth is fragile about init order, hence the care in AppDelegate.
Firebase backend. Firestore + Storage + Cloud Functions carry store/retrieve/tag/seal and the push notifications for Giver. One service (FirebaseService) owns all Firestore access, keeping the substrate boundary clean.
Swift, on-device ML. Native Swift for the app and on-device ML for tagging — private, free, instant, and the substrate for the place-tuned-models moat. On-device-first is a stance, not just an optimization (see ML Report and Science).
Traction
User growth, live from Firestore users.createdAt.
New users by week
messages collection once founder-read is on.Retention
How many users come back — the number investors care about most.
lastActiveAt we stamp on app open / record / play). The data model has createdAt but no lastActiveAt today.
To turn this on: (1) stamp
users/{uid}.lastActiveAt = serverTimestamp() on app foreground (iOS) — a 3-line change in AppDelegate/AuthViewModel; (2) deploy the founder-read rule. Then this page computes D1 / D7 / D30 retention and a cohort grid automatically.
Cohort retention (preview)
lastActiveAt is stamped.| Cohort | D1 | D7 | D30 |
|---|---|---|---|
| This week | — | — | — |
| Last week | — | — | — |
Audio ML — Roadmap & Reality Check
How Time Machine gets to hyper-specific listening — people, birds, songs, key/tempo, topics — and the place-tuning moat. Written plain.
Financials
Spend, runway, revenue, and docs — in one place.
Connect a source
💳 Card & bank — Plaid
Live card/bank transactions + balances. Plaid Link (OAuth) → a Vercel/Cloud Function holds the secret and stores transactions to Firestore. Then this page charts spend + runway.
📈 Revenue — Stripe
MRR, subscriptions, churn once billing is live. Stripe read-only key in a function → summarized here.
📄 Financial docs
finance/ Storage folder + founder-only Storage rules + an upload widget here. I can build the upload + list next; it just needs the rule deploy.Requests & comments
Leave a comment or an edit request — I read these and act on them. Think of it as a to-do line straight to the build.
New request
Open requests
Roadmap & shipped
What's live, what's next.
Shipped
- iOS core loop: record → seal (leave & return) → unlock → Play-the-space mix.
- Consolidated hexagon home (gesture center + corner commands), pink/green theme.
- Web: brochure + iOS-accurate Time Machine (map backdrop, seal, guest mode).
- YC demo account (
ycombinator) + text-a-link Giver invites. - Media ingest pipeline (camcorder/mp4/voice-memos → the TM database).
Next (my picks)
- ShazamKit song ID — native, scaffolded, one capability toggle from live.
- BirdNET bird species — the Merlin model, on-device, on-brand.
- Founder-read rule → this portal's locked tiles go live.
- Phone-auth on device — the last unproven pre-event item.
Behind the Curtain
Time Machine allows users to make and consume media only for and in the space they are in. This portal allows for you to simulate using Time Machine as if you were in these spaces. This is not a product demo, more of a database demo. Highly recommended to view on web or ios app these media in the spaces they were made on Time Machine. Mementōte…
to listen at that spot
This is a Time Machine, you can record something where you are, leave it where you are, return to it and play it back. Try it out here or on TestFlight.
Get Time Machine on TestFlight →Behind the Curtain →Science
I'm going to try to hone in my understanding of what the Time Machine could do to someone's cognition.
One's cognition is built in part by interaction between the hippocampus and the medial entorhinal cortex (MEC).
The MEC has grid cells, which in a 2d space, correspond to a grid of neurons that fire in patterns in reference to a grid that maps out one's own internal coordinate system.
The MEC is one's internal compass.
The MEC and its grid cells allow for one to keep track of their position in space in the absence of external cues—more on dead reckoning to come.
The MEC provides spatial phase and metric information (the "where" in abstract coordinates). The hippocampus binds this with sensory, emotional, and temporal information (the "what" and "when").
The hippocampus has place cells that do this semantic encoding of the grid cells.
The MEC uses these to create simulations and predict future events that help us survive.
This memory consolidation occurs during both sleep and awake states.
Awake states feature greater neuromodulators that focus on external cues.
Within the MEC, there is a complex adaptive cognition that occurs in regards to cognitive function and the pattern of neuronal firings within the grid cells.
During instability, times of newness, exploration, or surprise, the grid cells firing form a spontaneous emergent hexagonal pattern in order to stabilize.
During stability, there is a coherence in pattern as the mind moves.
Dead reckoning is one's ability through the MEC to maintain spatial and temporal orientation in the absence of external cues.
Stability is what is active during dead reckoning—one does not need external cues in order to know one's orientation in time and space.
Instability may occur if external cues cause instability in orientation.
A high level of cognitive function occurs with a push and pull of these times of stability and instability.
Dead reckoning allows for us to understand ourselves in time and space without external cues.
Generally what is thought of as an external cue that largely informs our understanding of time and space is from visual cues.
A walk to school or work is largely filled with place fields that are predictable and unmoving.
Our place field is largely made up of straight lines that don't seem to change.
This includes buildings, paths, roads, lights, phones, and other man made objects.
Our place fields are filled with things that are predictable and unmoving.
If this is the case, then we are frequently in stable states.
If we were to alter our external cues to create frequent place fields that are unpredictable and moving, would that make us more likely to experience unstable states thus allowing us to have a cognition that inches closer to the edge of chaos (adaptive and intelligent cognition)?
Would such an alteration make us more likely to be more creative, compassionate, and open to new experiences?
What would such an alteration look like?
Or maybe, what would such an alteration sound like?
Audition takes up space and time, but it is an external cue that we have a different relationship with and has different functioning than other senses.
We put audition as subordinate to how we orient ourselves in space and time.
Visual stimulus is our dominant point of reference.
Visual stimulus is our dominant world view.
Visual stimulus is often unmoving and predictable in man made places.
Sound can have straight lines, so to speak in terms of pitch, amplitude, rhythm, timbre and other forms of predictability and movement.
Most coffee shops sound the same. Most street corners sound the same. Often we have very similar conversations at work school—think charlie brown wa wa wah.
More importantly, sound is more ephemeral than visual cues.
Experience is ephemeral and phenomenal, but hard to differentiate from one second to the next.
Our experience is often lost in the stream of memory.
Memory cues are some of the few ways we are able to access our memory, think revisiting a childhood home, a hat from a party, songs, a photo, a journal entry, a letter, etc.
The hippocampus doesn't care what kind of space it's mapping—only that there's a coherent structure of relationships.
Sound can define that structure.
What if we made auditory stimuli for people in spaces that were moving and unpredictable, and yet held great semantic importance to the space it is inhabiting?
It seems this would make people's grid cells firing cause more adaptive and intelligent cognition.
How could this sound structure alter how that space is inhabited in the future?
How could this sound structure alter how that space's inhabitants understand the past?
How could this sound structure alter how people move through space?
The sound I have in mind is an aggregation of oral tradition in the forms of audio and text that may be compiled to be tied down to the space on earth that they occurred.
Instead of walking to school/work/life with the predictable and unmoving buildings and with the predictable bustling of the day/predictable music in headphones, one would have an auditory history of the grounds that they walk on that would be moving and unpredictable and yet tied down to a specific point in space.
The auditory history structure would be built by those who had walked and made noise in the same point in space.
The auditory future structure would be built by those that walk and make noise now in the same point in space—users input their now in order to preserve it for the future, and to gain access to the past.
The past would reverb and fill the void in sound.
People today are consumed by commercial existence and little connection with their history, communities, or their land.
This could help them know their history, know their communities, and know their land.
This could help them build their future, communities, and land better.
I envision this application to have similar effects to how maps change the way we understand ourselves in space—this application will try to change how we understand ourselves in time.
References
Burak, Y., & Fiete, I. R. Accurate Path Integration in Continuous Attractor Network Models of Grid Cells. Center for Brain Science, Harvard University, Cambridge. · Grossberg, S. A Neural Model of Intrinsic and Extrinsic Hippocampal Theta Rhythms: Anatomy, Neurophysiology, and Function. · Tolman, E. C. Cognitive Maps in Rats and Men. · Drieu, C., & Zugaro, M. Hippocampal Sequences During Exploration: Mechanisms and Functions. · Aronov, D., Nevers, R., & Tank, D. W. Mapping of a non-spatial dimension by the hippocampal–entorhinal circuit. · Kang, L., & DeWeese, M. R. Replay as wavefronts and theta sequences as bump oscillations in a grid cell attractor network.