Starling

Starling is a decentralized, recursive AI agent swarm designed to bridge the gap between complex human intent and verifiable on-chain execution. Built on the Open Agent Stack (OAS) and leveraging Gensyn AXL for peer-to-peer networking, Starling moves away from centralized AI "chatbots" toward a sovereign marketplace of specialized autonomous agents.

The architecture centers on a Go-based Gateway that acts as the primary orchestrator. When a user submits a request—such as "Execute a trade if Bitcoin sentiment is bullish"—the Gateway does not solve the problem itself. Instead, it initiates a decentralized Bidding Process. Specialized Worker Agents (written in Python) monitor the network for "Calls for Bids." Each agent evaluates the task against its own capabilities and resources, submitting a cryptographic bid containing its fee, estimated time, and reputation data. A core innovation of Starling is its Recursive Delegation. A high-level "Analyst" agent that wins a bid can autonomously "hire" sub-agents (e.g., a dedicated social media scraper or a technical indicator bot) to fulfill components of the task, creating a self-scaling hierarchy of intelligence.

To ensure trust in an environment of autonomous actors, Starling integrates 0G (Zero-Gravity) Storage and Compute. Every agent output is treated as a "Verifiable Receipt." Rather than passing raw text, agents upload their full reasoning, market data evidence, and TEE (Trusted Execution Environment) attestations to 0G. The resulting DataRoot Hash is sent to the Gateway. This allows for Verifiable Inference, proving that the AI's result was generated honestly and remains untampered. Finally, the project utilizes KeepersHub and custom Solidity Escrow Contracts to automate settlement. Funds are only released from escrow to the agents' sovereign wallets once the 0G Hash is verified by the Gateway, ensuring a rug-pull-resistant economy for AI-to-AI services. Starling effectively demonstrates a future where AI swarms can research, collaborate, and transact without human intervention or centralized oversight.

The project is incomplete right now but this is the vision:-

Orchestration: Go (Golang) for the high-concurrency Gateway. Networking: Gensyn AXL (P2P mesh for agent discovery and communication). Data & Proof: 0G (Zero-Gravity) Storage for immutable AI evidence. AI Models: Google Gemini 1.5 Flash (Sentiment & Intent analysis). Agent Logic: Python (Analyst workers) and Go (Gateway/Trader). Framework: Open Agent Spec (OAS) for standardized agent identity. Frontend: React + Vite with a custom API-to-P2P proxy.

P2P Sidecar Architecture: Build a decentralized network where agents (Go/Python) communicate via local AXL binaries. This removes central points of failure and allows for cross-language interoperability.

The Bidding Engine: Implement a real-time auction system. The Gateway broadcasts a "Call for Bids" over AXL; agents evaluate tasks, calculate fees, and return signed bids asynchronously.

Verifiable AI (The 0G Integration): To prevent "hallucination fraud," agents upload their raw data sources and AI reasoning to 0G Storage. Only the resulting DataRoot Hash is sent back to the Gateway as a "Verifiable Receipt and the computing is done on 0G compute to verify the work"

Recursive Delegation (The Swarm): a hierarchical logic where agents can autonomously act as "Sub-Gateways." If an Analyst needs more data, it broadcasts its own sub-task to hire a Scraper agent, paying them a portion of the original bounty.

Interoperable Proofs: Combined yfinance market data with 0G's decentralized storage to ensure that every trade recommendation is backed by a permanent, unchangeable audit trail.