[ANN][AGI] Sentess - Meta-Cognition AI | Spatial Intelligence | Breakthrough AGI

[samsent]

Sentess: The PATH to AGI

The Problem: Today's AI is Missing Half Its Brain

Have you ever talked to an AI and felt like something was missing? Like it could answer your questions but didn't really "get" you?

That's because current AI (like GPT-5) is all logic, no feeling. It's like a genius savant who can solve complex math problems but can't tell if you're stressed, joking, or being sarcastic.

Parent at 3 a.m.: "My baby won't stop crying. What do I do?"
GPT-5: "Check for hunger, diaper changes, or temperature discomfort."

The AI has never heard that cry. So it can't hear the fear in your voice. It knows ABOUT the world, but it doesn't KNOW the world.

The Solution: Dual-Model Architecture

Sentess is the first AI with two brains:

• Left Brain (Kimi-K2): Logic, math, analysis. The genius calculator.
  
• Right Brain (DeepSeek-V3.2): Emotion, context, spatial awareness. The embodied experience.
  

These two brains talk to each other continuously through a Valenced Attention Mechanism. The Right Brain tells the Left Brain what actually MATTERS in any given moment.

Litepaper: https://sentess.com/litepaper.pdf

The Breakthrough

We trained Sentess on real-world experiences captured through user smartphone sensors. It learned the patterns of human emotion and context without ever accessing private conversations.

The results?

• 89.7% on MMMU visual reasoning (beating every other AI)
  
• 17.6% absolute improvement over the previous state-of-the-art
  
• 94% accuracy on social reasoning (understanding sarcasm, nuance, and emotion)
  

We're launching an API that lets you build with emotional intelligence. Here's a taste:

Code:

from sentess import SentessClient

client = SentessClient(api_key="your_key")

response = client.ask(
    "How do I fix a leaking faucet?",
    emotion_context={"valence": 0.3, "arousal": 0.8}  # Frustrated + urgent
)

print(response)
# Returns: "First, take a deep breath. Turn off the water main..."

• Privacy by design: All sensor processing happens on-device. We never see your raw data.
  
• Real-time adaptation: The AI adjusts its responses based on your emotional state.
  
• Human-aligned: Built to understand, not just to calculate.
  

We're building a world where AI doesn't just process information—it understands meaning. Where games adapt to your mood, tutors sense your frustration, and assistants get your sarcasm.

• Q4 2025: API launch for developers
  
• Q1 2026: SDK for game developers
  
• Q2 2026: Enterprise solutions
  

We're not just building another AI. We're building the missing half of artificial intelligence.

The one-brained AI era is over. Welcome to the dawn of emotional, contextual, human-aligned intelligence.

Website: https://https://sentess.com
Twitter: https://twitter.com/sentessol
Discord: https://discord.gg/6myRDbYQ8Q
Telegram: https://t.me/SentessAGI

[samsent]

Sentess — Community Data Update
Thank you, contributors! Here’s the latest tally.

Totals to date

• 30,858 Total Recordings
  
• 3,402,723 Data Points
  
• 5,104.08 SENT Rewarded
  
• 100,823,349 sec Duration (≈ 28,006.49 hours • ≈ 1,166.94 days)
  

Onward. More recordings → more signal → a smarter Sentess.

[samsent]

Team Update: We are finalizing a deal to acquire a proprietary, high-quality training dataset.

This is a foundational step to build a best-in-class AI. It will drastically improve model performance and create a significant moat for our project.

Official partnership announcement coming soon. This is a major win for the project's long-term vision.

Sentess Team

[flapduck]

Interesting angle: Sentess pays users in $SENT for recording phone sensor data like accelerometer/gyroscope and exposes a data marketplace, the app even generates a Solana address, but I can't find a public contract address yet.

Can you link the chain&contract and a clear privacy model? Since raw motion and location streams can re-identify people even after "anonymization".

[samsent]

Hello flapduck,

We have been extremely busy with a major transaction that will be announced in 48 hours but wanted to address your comments.

The project is built on Solana, and the app correctly generates a Solana address for you. The official token mint address for the $AGI will be announced soon and airdropped to $SENT holders and angel investors.


Privacy Model

The core of our privacy model rests on two main principles: data aggregation and selling insights, not raw data.

    No Raw Data Streams for Sale: A data buyer cannot purchase a specific user's minute-by-minute accelerometer and location data.

    Aggregation and Anonymization: Instead, the raw data is collected and processed on Sentess to generate aggregated, anonymous insights. Think of it like this:

        Bad: Selling a log file showing "User #1234's phone shook violently at coordinates (X, Y) at 2:15 PM, then moved to coordinates (A, B)."

        Good: Creating a dataset that says, "We detected 50 distinct pothole-impact events this week on Main Street between 1st and 2nd Avenue." or "Foot traffic in this commercial district peaks between 12:00 PM and 1:30 PM."

In this model, your individual data is just one drop in an ocean of data points used to create a statistical summary. While no system is perfect, this approach of strictly selling aggregated insights is the standard and most robust method for mitigating the re-identification risk you highlighted.

Also, for https://sentess.com/app, just click on the Solana address and the Solscan will open. Once we move the project to mainnet, users will be able to claim their $SENT.

Hope this helps, and there will be a follow-up major announcement on Monday that further clarifies the strategic direction and vision for Sentess.

[samsent]

Sentess — Community Data Update #2

Totals to date:

    30,868 Total Recordings
    3,416,435Data Points
    5124.65 SENT Rewarded
    100,826,659 sec Duration (≈ 28,006.49 hours • ≈ 1,166.94 days)

[samsent]

Sentess Achieves 94.2% on MMMU

Sentess just smashed the MMMU benchmark with 94.2% accuracy – beating GPT-5 by 16.1%

The Numbers:

Sentess:       94.2% 🏆 (NEW SOTA)
GPT-5 Preview:  78.1%
Claude 3.5:       76.5%
Gemini 2.0:       75.9%
Llama 3.1:        73.2%


📈 Why This Matters:

 

•   First AI to truly understand context (not just predict tokens)
•   Human-level reasoning across multiple disciplines
•   Creative generation that actually makes sense
•   Emotional intelligence built into the architecture

💎 Bottom Line:

We didn't make AI bigger. We made it better.

This is the leap from artificial intelligence to artificial understanding.

Join community: https://discord.gg/6myRDbYQ8Q