← All cases
AI Memory SaaSData Processing · Personality Modeling · Content Generation

etersoul.me

95% of the Memory-Processing Pipeline Runs Fully Automatically on AI Employees

EterSoul (your Second Life) is your mirror in the digital world—through native multimodal sensors it continuously absorbs the moments of your life, building in the cloud a digital being that carries your thought patterns and personality, able to converse with you privately and even chat with friends on your behalf when you're busy.

95%

Memory-Processing Automation

Classification and modeling done with zero human input

Memory-Processing Throughput

vs. a pure rules-based system

–82%

Personality-Modeling Latency

Initial blueprint for new users within 24h

THE CHALLENGE

The Challenge

EterSoul's core value lies in the quality of memory—the voice notes, photos, and long-form text users upload must be correctly classified, emotion-tagged, and relationship-mapped to build an authentic digital soul. The early rules engine had low throughput and coarse classification; with such varied content types (dialect voice notes, handwritten scans, cross-language material), rules struggled to cover every scenario, leaving the memory store uneven in quality.

THE SOLUTION

How the AI Employees Step In

EterSoul migrated its entire core memory-processing pipeline to a RixyAgent-powered AI employee system: a "Memory Curator" handles multimodal classification and tagging, a "Personality Analyst" extracts a personality blueprint from conversations and content, and a "Dream Cycle Curator" runs at night to weave fragmented memories into structured weekly-journal summaries pushed to users for confirmation each morning. The whole pipeline runs continuously around the clock, cutting processing latency from hours to minutes.

USE CASES

Three AI Employee Workflows

01

Multimodal Memory Classification

Multimodal Memory Classification

The AI employee ingests the voice notes, images, and text fragments users upload and automatically identifies content type (journal, conversation, event record), emotional tone (positive/neutral/negative), timeline placement, and relationships to existing memories. It handles mixed Mandarin, Cantonese, and English content, with classification accuracy far above the rules engine—providing high-quality raw material for downstream personality modeling.

Before

Rules engine, 61% accuracy

After

AI employee, 91% accuracy

02

Personality Blueprint Extraction

Personality Blueprint Extraction

The AI employee continuously analyzes the user's memory stream—distilling a "personality blueprint" from recurring vocabulary preferences, emotional patterns, and expressions of values, including an interest matrix, communication style, emotional tendencies, and a social-relationship graph. An initial blueprint is built within 24 hours of a new user joining, then updated incrementally each day. The blueprint directly drives the digital soul's dialogue generation, shaping the tone and perspective of its replies.

Before

New users could only experience it after 72h

After

Initial blueprint within 24h

03

Dream Cycle Nightly Summarization

Dream Cycle Nightly Summarization

Each night the AI employee automatically scans the day's memory fragments and generates a "life weekly journal" structured along three axes—event, emotion, reflection—paired with key memory snippets and AI narration. It is pushed to the user the next morning for confirmation and correction; confirmed summaries enter the long-term memory store, continuously reinforcing the digital soul's sense of authenticity.

Before

No regular curation, memories scattered

After

Generated daily, 78% user confirmation rate

The hard part of building a digital soul isn't the AI's reasoning ability—it's the depth of understanding when processing human memory. RixyAgent's AI employees have persistent memory and context awareness, letting us truly deliver on the product vision of "the more you use it, the better it knows you."

EterSoul Team

Digital Soul Platform · etersoul.me

GET STARTED

Unlock x

Solve for x — with a team of AI employees