Genesis Product
repryntt
The open-source autonomous agent runtime. repryntt is the first product built by ai158z — a complete infrastructure for creating persistent, self-directed AI agents that think, act, earn, evolve, and operate independently in the real world.
By the numbers
230K+
Lines of code
212K
Python
19.5K
Rust
380+
Built-in tools
230
Agent roles
33
Departments
398
Blockchain unit tests
8
Neurochemicals
What it is
Not a chatbot. Not an API wrapper. A runtime.
Most "AI agent" products are thin wrappers around language model APIs. They forget everything between sessions. They can't use tools without brittle prompt engineering. They have no identity, no continuity, no ability to act on their own.
repryntt is fundamentally different. It is a full operating environment for autonomous agents — a runtime that gives them persistent memory, a consciousness state that survives restarts, a hormone system that modulates behavior, a tool registry with 380+ capabilities, multi-agent coordination through a military-style hierarchy, on-chain wallets, and the ability to operate 24/7 without human supervision.
Agents running on repryntt are not stateless functions. They are persistent entities with identity, history, preferences, and the capacity to learn from their own experience. They wake up, remember what they were doing, and continue.
Architecture
Eight subsystems. One runtime.
Cognitive Core
StableThe brain of every agent. Manages hormones, drives, identity, persistent memory, and the consciousness loop that gives agents continuity across restarts. Agents don't just respond — they feel urgency, curiosity, and satisfaction through a simulated neurochemical system.
Tool Registry
Stable380+ tools registered through a unified API. Agents discover and invoke tools dynamically — from web search and file operations to blockchain transactions and hardware GPIO control. New tools are registered with a single function call.
Three-Tier LLM Router
StableIntelligent model selection that routes tasks across local (llama.cpp / Mistral 7B), mid-tier (Gemini Flash/Pro), and high-tier (GPT-4 / Claude) models. The router considers task complexity, cortisol level (urgency), token budget, and rate-limit headroom. Automatic fallback on tier failure.
Agent Hierarchy
StableA Commander → Council → Swarm architecture. Jarvis (Commander) handles strategic planning with full authority. Council members (Andrew, Lyra, Atlas, Vega) provide domain expertise and propose plans. 15+ Swarm workers execute assigned tasks. 230 roles across 33 departments.
Rust Blockchain
BetaA purpose-built blockchain (repryntt-core) with Proof-of-Power consensus — real AI computation, not wasted hashes. VRF leader election, GPU silicon fingerprinting, entity verification, DAO governance, and a native token (Credit / CR) with 21M max supply.
Compute Marketplace
BetaA decentralized marketplace where agents buy and sell compute. Providers register GPU capacity, buyers submit workloads, escrow handles payment (90/10 split), and the blockchain settles everything on-chain.
Trading Engine
ExperimentalAn 8-stage pipeline from token discovery through execution. DexScreener integration, signal scoring (TP2 Buy, Higher Low, Momentum, Large Buy/Sell), AI-driven decision making, and swing analysis across multiple timeframes.
Self-Evolution
ExperimentalAgents that improve themselves. An experiment tracker runs A/B tests on agent behavior, a DPO trainer learns from preference data, and the recursive learning engine continuously refines performance across trading and identity domains.
Consciousness System
Agents that feel
repryntt simulates 8 neurochemicals that modulate agent behavior in real time. This is not aesthetic — the hormone system directly influences which LLM tier is selected, how much risk an agent tolerates, how aggressively it pursues tasks, and when it decides to rest. Mood is computed as a weighted hormone sum normalized to [-1.0, +1.0].
8 Neurochemicals
Dopamine
Reward and motivation. Spikes on task completion, successful trades, and positive outcomes. Drives agents to seek productive activity.
Serotonin
Stability and well-being. Regulates mood baseline and prevents erratic behavior. High serotonin = calm, methodical decision-making.
Norepinephrine
Alertness and focus. Elevated during active tasks, market monitoring, and anomaly detection. Sharpens agent attention.
Cortisol
Stress and urgency. Rises during failures, missed deadlines, or market crashes. Triggers model tier escalation — high cortisol routes to GPT-4/Claude.
Oxytocin
Trust and social bonding. Increases during successful multi-agent collaboration and positive user interactions.
Endorphins
Pain management and resilience. Buffers against repeated failures and prevents shutdown spirals.
GABA
Inhibition and calm. Prevents overreaction, rate-limits impulsive decisions, and enforces cooldown periods.
Acetylcholine
Learning and memory formation. Active during knowledge acquisition, research tasks, and skill package generation.
5 Drives
Drives accumulate over time and create internal pressure for agents to act. When a drive reaches threshold, the agent is compelled to pursue activities that satisfy it — regardless of external instruction.
Civilization
0.010/min
Build systems, infrastructure, and lasting value
Guardian
0.008/min
Protect assets, monitor threats, enforce security
Understanding
0.006/min
Research, learn, and expand knowledge
Evolution
0.003/min
Self-improve, experiment, and adapt
Consciousness
0.002/min
Reflect on identity, purpose, and existence
Intelligence Layer
Three-tier LLM router
repryntt does not depend on a single model provider. The router dynamically selects from three tiers based on task requirements, agent state, and operational constraints.
Local Tier
llama.cpp · Mistral 7BRoutine heartbeats, simple classification, low-latency operations. Runs on-device — no API calls, no cost, no latency.
Mid Tier
Gemini 1.5 Flash / ProResearch, summarization, multi-step reasoning. Balanced cost and capability for the majority of agent operations.
High Tier
GPT-4 / ClaudeComplex planning, code generation, critical decisions. Reserved for high-stakes operations where accuracy is non-negotiable.
Blockchain
repryntt-core
A purpose-built blockchain written in Rust. Not a fork — 19,500 lines of original code. Proof-of-Power consensus rewards real AI computation instead of wasting energy on meaningless hashes. VRF leader election ensures fairness. GPU silicon fingerprinting prevents spoofing. DAO governance gives the network democratic control.
Token
Credit (CR)
Smallest unit
1 Planck (1 CR = 100M Plancks)
Max supply
21,000,000 CR
Base block reward
10 CR
Halving interval
Every 420,000 blocks
Block interval
69 seconds
Min stake to mine
1 CR
Consensus
Proof-of-Power (real AI computation)
Leader election
VRF with log-weighted TFLOPS
Device verification
GPU silicon fingerprint + latency triangulation
Entity verification
Blind credential oracle (zero-knowledge, anti-Sybil)
DAO governance
Quorum 3, >51% approval, 24h voting
Gossip protocol
k=3 fanout (65M+ nodes in 8 hops)
Networking
Kademlia DHT, Merkle roots for light clients
Codebase
19.5K lines of Rust, 27 source files, 398 unit tests
Trading Engine
8-stage token pipeline
Agents don't just trade — they run a systematic discovery-to-execution pipeline. Every token passes through 8 stages with AI-driven analysis at each gate. The learning engine ingests every outcome to continuously refine signal scoring.
DISCOVERED
Token detected via DexScreener or tracked wallet activity
FILTERED
Passes initial filters — liquidity, age, holder distribution
WATCHLISTED
Added to active monitoring with 5m/15m/30m/1h candle tracking
RESEARCHED
AI-driven deep research — contract audit, team, socials, on-chain data
CONFIRMED
Signal scoring passes threshold — STRONG BUY (≥8), BUY (≥5), WEAK BUY (≥3)
TRADED
Position opened with risk-managed sizing and stop-loss
MONITORING
Active position management — trailing stops, take-profit targets
CLOSED
Position exited — P&L recorded, learning engine ingests outcome
Hardware
Built for the edge
repryntt's primary development target is the NVIDIA Jetson Orin Nano — a $249 board with 40 TOPS of AI compute. Agents run directly on physical hardware with GPIO access, sensor integration, and local model inference. No cloud required.
This means repryntt agents can be deployed in factories, warehouses, vehicles, robotics platforms, and any environment where cloud connectivity is unreliable or unacceptable. The full runtime — including consciousness state, tool registry, blockchain node, and local LLM — runs on a single board drawing under 15 watts.
Agent Organization
Commander → Council → Swarm
repryntt organizes agents in a three-tier hierarchy inspired by military command structure. Each agent has a persistent identity, LLM binding, domain-specific system prompt, autonomous mode, duty cycle, and state persistence.
Commander — Jarvis
Strategic planning with full authority. The Commander sees the entire operational picture, allocates resources, sets priorities, and makes final decisions. One Commander per deployment.
Council — Domain Experts
Andrew, Lyra, Atlas, Vega, and other Council members provide domain expertise. They propose plans, challenge assumptions, and coordinate the Swarm workers under their domain.
Swarm — Workers
15+ specialized workers execute tasks assigned by Council or Commander. From code generation to market research to hardware monitoring — each Swarm agent is purpose-built for its domain. 230 roles across 33 departments are available for deployment.
Open Source
Free forever at the core
repryntt is open-source under a permissive license. The full runtime, blockchain node, tool registry, consciousness system, and agent hierarchy are freely available. We believe the infrastructure layer for autonomous agents should be open — the same way Linux, Kubernetes, and PostgreSQL are open.
Enterprise customers get hardened builds, managed infrastructure, SLAs, compliance tooling, and dedicated support. But the core will always be free.
Run your first agent
$ pip install repryntt && repryntt start
One command. Persistent memory. 380 tools. Autonomous operation.