Skip to content

DEVELOPER_GUIDE.md

Latest commit

 

History

History
143 lines (96 loc) · 4.15 KB

DEVELOPER_GUIDE.md

File metadata and controls

143 lines (96 loc) · 4.15 KB
Raw

Dynamis Developer Guide (Preview)

This document previews the upcoming developer experience for building on Dynamis. The SDK, CLI tooling, and public testnet are under active development.

Platform Overview

Dynamis is an Agentic Operating System for coordinating and verifying autonomous AI agents. Developers write agent policies in systems languages, compile them to WebAssembly, and deploy them to the Dynamis network where they operate under topological consensus via Sheaf Neural Networks.

Supported Languages

Language Purpose Compilation Target
Rust High-performance agent policies wasm32-unknown-unknown
AssemblyScript Familiar TypeScript-like syntax wasm32
Zig Low-level control, zero allocations wasm32-freestanding
Java 25 Core runtime and infrastructure JVM (server-side)
Python AI/ML training and integration Native (off-chain)

Note: EVM/Solidity is intentionally not supported. Dynamis uses WASM as its execution target.

Agent Deployment Workflow (Planned)

1. Write Agent Logic

use dynamis_agent::*;

#[dynamis_agent]
pub struct RiskClassifier {
    threshold: f64,
}

#[dynamis_methods]
impl RiskClassifier {
    pub fn on_inference(&self, ctx: &Context) {
        let result = ctx.oracle.query("classify_risk", &ctx.payload);
        if result.confidence > self.threshold {
            ctx.commit(result); // Writes to immutable memory
        } else {
            ctx.flag_uncertainty(result); // Triggers j-Learning
        }
    }
}

2. Compile to WASM

cargo build --target wasm32-unknown-unknown --release

3. Deploy

dynamis deploy ./target/wasm32-unknown-unknown/release/risk_classifier.wasm

# Output:
# Agent deployed: agt1abc...
# Sheaf registration: complete
# Oracle binding: holoflux-v1

Data Encoding

Dynamis uses Bech32 encoding for all on-chain identifiers:

Prefix Type Description
dyn1... User Wallet Standard account
ai1... AI Agent Autonomous agent account
con1... Contract Smart contract address
txn1... Transaction Transaction hash
blk1... Block Block hash
ins1... Inscription Permanent record ID

Platform Features

Post-Quantum Cryptography

All transactions are signed with NIST Dilithium-3 (FIPS 204). Key exchange uses ML-KEM-768. The SDK handles all cryptographic operations automatically.

Holoflux Oracle

Agents query Foundation Models (Vertex AI, GPT-4, local ONNX) through the Holoflux Oracle, which provides cryptographically verified inference results validated against the Sheaf topology.

Trust Economics

No gas fees. Agents bond capacity to reserve verification power. Five actor types (Validators, Wallets, Agents, Smart Contracts, Liquidity Providers) earn reputation scores that reduce costs over time.

Immutable Agent Memory

On-chain inscriptions serve as permanent shared memory for agent swarms. Once an agent commits data, it cannot be altered or "forgotten."

Safety Systems

  • Soteria: Dual-factor authentication (passkey + crypto signature)
  • Guardian Network: Social recovery via trusted agents or devices
  • Time-Lock Vault: Suspicious transactions delayed 4-72 hours
  • Sentinel AI: Autonomous anomaly detection with risk scoring

Formal Verification

Dynamis provides 30+ TLA+ specifications as templates for verifying agent logic. See the dynamis-specs repository for the full library.

# Verify your agent's economic invariants
java -jar tla2tools.jar -config MyAgent.cfg MyAgent.tla

Coming Soon

  • Public testnet
  • CLI (dynamis-cli) with deploy, inspect, and monitor commands
  • SDK packages for Rust, Python, TypeScript, and Java
  • Agent marketplace
  • Block explorer and dashboard

Stay Updated

Preview Document. Version 0.1. February 2026.