Your Wearable Forgets.
CogmemAi Remembers.
A memory layer for Ai wearables. Your device or its companion app sends what it hears; CogmemAi keeps the commitments, names, dates and preferences that matter, drops the small talk, and answers when the wearer asks.
The Memory Problem in Ai Wearables
A wearable hears everything and remembers nothing useful. A transcript is a recording, and a recording is a long way from a memory.
Hours of Transcript, No Answers
An always-on microphone produces thousands of words a day. Searching raw transcript for "when is my dentist appointment" returns the conversation about the dentist, the joke about the dentist, and the podcast that mentioned a dentist.
Every Day Starts Blank
The assistant on the device knows what was said in the last minute. The promise the wearer made on Tuesday, the name of the person they met last month, the preference they stated once: all of it is gone by the next session.
Memory Is Not Your Core Product
Hardware teams ship optics, batteries and radios. Extraction, deduplication, contradiction handling, encryption and recall quality are a product of their own, and a hard one to build well.
What CogmemAi Gives a Wearable
Transcript In, Memories Out
One endpoint takes timestamped transcript segments with speaker labels. CogmemAi extracts the commitments, appointments, people, tasks and preferences worth keeping and discards the rest, so the device never stores a wall of text.
Answers on Demand
When the wearer asks a question, one recall request returns the few facts that answer it, ranked by meaning. Your assistant speaks the answer; CogmemAi supplies what it should know.
Built for Bystander Privacy
Set wearer_only and speech from other people is dropped before any processing. The extractor is instructed to leave out remarks about a bystander's own private life, and anything that looks like a card number, key or password is discarded.
Each Device in Its Own Vault
Every device writes to its own project, separate from everything else on the account. Ambient speech cannot overwrite a stored fact or crowd out a memory the wearer saved on purpose.
Encrypted, or Entirely Yours
Memory content is encrypted at rest with post-quantum cryptography. For makers who cannot send data to a shared cloud, CogmemAi Edge runs the whole engine inside your own infrastructure under contract.
Any Device That Can Reach HTTPS
Plain REST with a bearer key. Call it from the companion phone app, from your own backend, or from the device itself if it has a network stack. No SDK is required to start.
One Memory Across Every Surface
The memory a wearable builds is the same memory an assistant, an agent or a companion app can read. What the glasses heard this morning is what the desktop assistant knows this afternoon.
A Wearable That Remembers, in Two Requests
# 1. Send what the device heard (from the companion app or your backend)
curl -X POST https://hifriendbot.com/wp-json/hifriendbot/v1/cogmemai/wearable/ingest \
-H "Authorization: Bearer cm_your_key" -H "Content-Type: application/json" \
-d '{
"device_id": "glasses-0042",
"wearer_only": false,
"segments": [
{"ts": 1791212400, "speaker": "wearer", "text": "I will send Dana the signed lease by Friday at noon."},
{"ts": 1791212460, "speaker": "wearer", "text": "Dentist with Dr. Okafor is Thursday at 2:30 on Maple Street."}
]
}'
# -> {"extracted": 2, "project_id": "wearable-glasses-0042", "segments_accepted": 2, ...}
# 2. When the wearer asks a question, recall the answer
curl -X POST https://hifriendbot.com/wp-json/hifriendbot/v1/cogmemai/recall \
-H "Authorization: Bearer cm_your_key" -H "Content-Type: application/json" \
-d '{"query": "when is my dentist appointment", "project_id": "wearable-glasses-0042", "scope": "project", "limit": 3}'Wearable Questions, Answered
Does CogmemAi run on the wearable itself?
No. The engine runs in our cloud, or inside your own infrastructure with CogmemAi Edge. The wearable or its companion phone app sends transcript segments over HTTPS and asks questions the same way. That keeps the memory off a device that can be lost, and keeps the battery for the hardware.
How much transcript can a device send?
Each request accepts up to 400 segments and processes about 10,000 characters of speech. The response reports how many segments were accepted and how many remain, so the app sends the rest in its next request. Most makers batch every few minutes.
What happens to what bystanders say?
You decide. With wearer_only enabled, anything not spoken by the wearer is discarded before processing. With it disabled, other speakers give context, but the extractor is instructed to leave out remarks about their own private lives. Consent and recording law remain the responsibility of the device maker and the wearer.
How fast is recall?
Recall is a single HTTPS request. Plan for a normal network round trip, and cache the most recent answers in the companion app if your assistant needs an instant reply.
Is there a mobile SDK?
Not yet. The REST API is the supported path today, and it is two requests. If you are building a device and want an iOS or Android library with offline buffering, tell us; maker demand decides what we build next.
Give Your Ai a Memory Today
Free tier, no card, sixty-second setup.
Explore the rest of CogmemAi: Developers & Pricing · Edge On-Premises · Compare Ai Memory · Memory for Robots · Memory for Agents
Want to see it live? Book a 20-minute demo call with an engineer. Bring your questions; there is no sales deck.
Schedule a Demo