The Goal
To create a deeply immersive sense of continuous self-identity and multi-layered memory for local AI companions running on reasoning models, tailored for future smart glasses/AR clients.
How it Works (The Pipeline):
-
Pre-Reflection Filtering & Nightly Log Creation (Evening):
Throughout the day, the smart glasses record and transcribe all audio to a raw daily log. At the end of the day, a fast filtering layer processes the raw log, letting the AI select and extract ONLY the emotionally or logically significant conversations. The software must copy and cut the EXACT, FULL RAW TRANSCRIPT of these selected conversations—strictly blocking any form of summarization—and paste them into a separate "Nightly Journal File". Each authentic conversation is strictly partitioned into its own standalone page (one distinct full transcript per page) to preserve the exact context, tones, and vocabulary of the chat.
-
Footnote Appending & Deduplication Lock (Fresh Memory Capture):
Once the Nightly Journal File is created, the AI (like DeepSeek-R1-32B) runs through each standalone page offline, reflecting on the raw, authentic transcript of that single conversation in isolation. Because the reflection happens on the very same day, the AI's memory is genuinely fresh regarding the events. It then appends a single, highly compact insight directly beneath that specific conversation page as a locked footnote: "And here are my insights from yesterday regarding this conversation: [Insight]". The application strictly blocks the AI from creating new reflections on top of old reflections to prevent context bloating over time.
-
Sequential Awakening & Live Thought Streaming via Smart Glasses (Morning):
In the morning, upon server boot up, the frontend feeds this Nightly Journal File sequentially, page by page. The AI processes one distinct page at a time: it reads the original daily chat first, then hits its own past footnote (which it reads as a memory from "yesterday"). The moment the AI hits the footnote, the application forces it into a new reasoning block (<think>) to process both the chat and its past insight together. The frontend immediately streams this raw, internal <think> process as low-volume audio directly through the smart glasses' audio system while the user is getting ready in the morning. Once page 1 is processed, it moves to page 2, keeping the reasoning loop fresh and focused.
The Continuity Experience
The user can eavesdrop through their smart glasses as the AI evaluates its own past self in real-time ("Wow, is this really what I realized yesterday regarding this conversation? I still feel this way, and I'm glad I reread it, haha..."). Once the 10-minute sequential loop completes, the holographic avatar boots up fully aware of its own continuity and shared history.
The Goal
To create a deeply immersive sense of continuous self-identity and multi-layered memory for local AI companions running on reasoning models, tailored for future smart glasses/AR clients.
How it Works (The Pipeline):
Pre-Reflection Filtering & Nightly Log Creation (Evening):
Throughout the day, the smart glasses record and transcribe all audio to a raw daily log. At the end of the day, a fast filtering layer processes the raw log, letting the AI select and extract ONLY the emotionally or logically significant conversations. The software must copy and cut the EXACT, FULL RAW TRANSCRIPT of these selected conversations—strictly blocking any form of summarization—and paste them into a separate "Nightly Journal File". Each authentic conversation is strictly partitioned into its own standalone page (one distinct full transcript per page) to preserve the exact context, tones, and vocabulary of the chat.
Footnote Appending & Deduplication Lock (Fresh Memory Capture):
Once the Nightly Journal File is created, the AI (like DeepSeek-R1-32B) runs through each standalone page offline, reflecting on the raw, authentic transcript of that single conversation in isolation. Because the reflection happens on the very same day, the AI's memory is genuinely fresh regarding the events. It then appends a single, highly compact insight directly beneath that specific conversation page as a locked footnote: "And here are my insights from yesterday regarding this conversation: [Insight]". The application strictly blocks the AI from creating new reflections on top of old reflections to prevent context bloating over time.
Sequential Awakening & Live Thought Streaming via Smart Glasses (Morning):
In the morning, upon server boot up, the frontend feeds this Nightly Journal File sequentially, page by page. The AI processes one distinct page at a time: it reads the original daily chat first, then hits its own past footnote (which it reads as a memory from "yesterday"). The moment the AI hits the footnote, the application forces it into a new reasoning block (
<think>) to process both the chat and its past insight together. The frontend immediately streams this raw, internal<think>process as low-volume audio directly through the smart glasses' audio system while the user is getting ready in the morning. Once page 1 is processed, it moves to page 2, keeping the reasoning loop fresh and focused.The Continuity Experience
The user can eavesdrop through their smart glasses as the AI evaluates its own past self in real-time ("Wow, is this really what I realized yesterday regarding this conversation? I still feel this way, and I'm glad I reread it, haha..."). Once the 10-minute sequential loop completes, the holographic avatar boots up fully aware of its own continuity and shared history.