Skip to content
@SivletLabs

SivletLabs

System-1 Decision Models & Evaluation Infrastructure. Typed decisions, calibrated probabilities, pay-per-call via x402.
README.md
SivletLabs

⚡ SivletLabs

System-1 Decision Models & Evaluation Infrastructure for Autonomous Agents

Website Gymnasium Compliant x402 Micropayments Tokenomics

Typed decisions. Calibrated probabilities. Pay-per-call via x402. Protocol revenue buys back & burns the token.

🏛️ What We Build

Most autonomous AI agents suffer from the "Autoregressive Trap": using 70B+ chat LLMs for discrete decisions (routing, tool selection, guardrails, DOM clicks). This incurs 800ms–2500ms latency, high cost, and uncalibrated probabilities.

SivletLabs builds the cognitive dual-system stack for agents:

                  ┌─────────────────────────────────────────────────────────┐
                  │                 Autonomous Agent System                 │
                  └───────────────┬─────────────────────────┬───────────────┘
                                  │                         │
                    System-1 (Spinal Reflex)    System-2 (Deliberation)
                                  │                         │
                  ┌───────────────▼─────────┐   ┌───────────▼───────────────┐
                  │    Sivlet-Jev Model     │   │   Heavy Autoregressive    │
                  │  (Single-Pass Forward)  │   │       Reasoning LLM       │
                  │   Latency: ~12ms        │   │    Latency: 800ms-3000ms  │
                  └───────────────┬─────────┘   └───────────────────────────┘
                                  │
                  ┌───────────────▼─────────────────────────────────────────┐
                  │  jev-eval: Gymnasium RL Benchmark & Calibration Score   │
                  │  Strictly Proper Scoring (Log / Negative Brier / Spher) │
                  └───────────────┬─────────────────────────────────────────┘
                                  │
                  ┌───────────────▼─────────────────────────────────────────┐
                  │  x402 Micropayment Gateway (Pay-per-decision in USDC)   │
                  └───────────────┬─────────────────────────────────────────┘
                                  │
                  ┌───────────────▼─────────────────────────────────────────┐
                  │  100% Net Revenue -> TWAP Buyback & Burn ($SIVLET)      │
                  └─────────────────────────────────────────────────────────┘

🔬 Core Ecosystem Projects

Repository / Module Description Status
contracts Production smart contracts: BuybackBurnEngine.sol, Clanker launch integration, and Uniswap v3 multi-hop buyback router. Audited & Tested
x402-gateway Edge-native HTTP 402 multi-channel failover gateway with dual-currency (USDC + $SIVLET) settlement. Live on Edge
jev-eval Gymnasium-compatible RL simulation, GAE buffer engine, and evaluation benchmark for discrete System-1 policies. v0.2.0 Active
jev-local Apple Silicon MLX local inference engine with KV-cache truncation (trim_prompt_cache), providing 24.3× speedup. Active
sivletlabs.github.io Official portal, interactive simulation sandbox, live telemetry, and technical litepaper. Live

📈 Academic Benchmark Highlights (8x8 Maze Navigation)

Evaluated under sequential MDP $\mathcal{M} = \langle \mathcal{S}, \mathcal{A}, \mathcal{P}, \mathcal{R}, \rho_0, \gamma \rangle$ ($N=10, T=25, \gamma = 0.99$):

Policy Candidate Win Rate Cumulative Return ($G$) Discounted Return ($G_{0.99}$) Mean Horizon ($T$) Latency ($P_{50}$)
HttpPolicy (jev-latest) 100.0% +9.84 ± 0.00 +8.51 16.0 steps 12.5 ms
OraclePolicy (Theoretical Ceiling) 100.0% +9.84 ± 0.00 +8.51 16.0 steps 0.05 ms
RandomPolicy (Uniform Baseline) 0.0% -0.20 ± 0.00 -0.18 20.0 (Timeout) 0.02 ms

Proven: With anti-oscillation state augmentation and strictly proper scoring, Sivlet-Jev matches the theoretical ground-truth oracle path length and reward ceiling.

🔄 Closed-Loop Protocol Economics & Fair Token Launch

           [ Agent / Developer Client ]
                         │
              ┌──────────┴──────────┐
              │  Pay with $SIVLET   │  Pay with USDC
              │  (20% Discount)     │  (Standard Fee)
              ▼                     ▼
        [ 0x...dEaD ]         [ Protocol Treasury ]
        (Direct Burn Sink)    (BuybackBurnEngine.sol)
                                    │
                       ┌────────────┴────────────┐
                       │ Uniswap v3 Multi-Hop    │
                       │ USDC ─[0.05%]─► WETH    │
                       │   │                     │
                       │   └─[1.00%]──► $SIVLET  │
                       ▼                         │
                 [ 0x...dEaD ] ◄─────────────────┘
              (Permanent Burn Sink)
  1. Clanker Fair Launch on Base: Fixed 1B supply launched fairly via @clanker on Warpcast / Base L2 with Uniswap v3 WETH pool.
  2. Dual-Currency x402 Micropayments: Agents pay in USDC (standard) or native $SIVLET (20% utility discount, burned directly on use).
  3. Autonomous Multi-Hop Buyback: 100% of USDC net protocol revenue in the Treasury triggers atomic Uniswap v3 swaps (USDC -> WETH -> $SIVLET) straight to 0x...dEaD.
  4. Decentralized Keeper Incentive: Anyone calling executeBuybackAndBurn() is rewarded with 0.5% (50 bps) of the transaction in USDC to cover gas.

🔗 Quick Links & Resources

© 2026 SivletLabs. Open-source decision intelligence for autonomous agent ecosystems.

Popular repositories Loading

  1. .github .github Public

    Organization profile and standards for SivletLabs

  2. sivletlabs.github.io sivletlabs.github.io Public

    SivletLabs Official Portal - System-1 Decision Models & Evaluation Infrastructure

    HTML

  3. contracts contracts Public

    Production smart contracts for SivletLabs System-1 Decision Network, Clanker integration, and x402 BuybackBurnEngine on Base L2

    JavaScript

Repositories

Showing 3 of 3 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…