SwarmEx

SwarmEx: Verifiable Agentic Intelligence Secured by the 0G Stack

SwarmEx pioneers the next generation of autonomous on-chain intelligence by building entirely on the 0G decentralized AI and Data Availability (DA) stack. Moving away from centralized, black-box LLM providers, SwarmEx is engineered to ensure that every layer of agentic decision-making is cryptographically verifiable, censorship-resistant, and permanently auditable. By deeply integrating 0G Compute for decentralized inference and 0G Storage for high-throughput execution provenance, SwarmEx establishes a trustless "Proof-of-Execution" pipeline for complex multi-agent swarms.

Decentralized Inference via 0G Compute

Traditional AI agents rely on centralized APIs, creating a massive trust deficit when handling on-chain execution. SwarmEx solves this by routing all critical agent logic through 0G Compute.

* Verifiable State Transitions: The multi-agent orchestrator utilizes 0G Compute to process complex natural language intents, dynamically calculate execution parameters, and evaluate real-time risk.

* Trustless AI Execution: By leveraging 0G’s decentralized serving infrastructure, we ensure that the inference determining the swarm's behavior is immune to centralized downtime, censorship, or silent model deprecation. The intelligence driving the swarm is mathematically verifiable and distributed.

Immutable Provenance with 0G Storage

Multi-agent swarms generate massive amounts of interaction data, intermediate states, and reasoning logs. To solve the AI "black box" problem, SwarmEx utilizes the unparalleled throughput of the 0G Storage network to anchor the entire lifecycle of an execution.

* High-Bandwidth DA Integration: Every inference output, state change, and metadata packet generated during a session is packaged and submitted directly to 0G Storage.

* Cryptographic Audit Trails: We leverage 0G’s highly scalable Data Availability layer to create a permanent, tamper-proof audit log. Every action taken by the swarm returns a root hash from the 0G Storage network, creating an immutable "Proof-of-Execution" that guarantees absolute transparency for all agentic decisions.

Bridging Off-Chain Intelligence with On-Chain Security

SwarmEx demonstrates the ultimate utility of the 0G ecosystem by bridging off-chain decentralized AI with strict on-chain validation mechanisms.

* AI-Driven Risk Attestation: A specialized Risk Guard agent uses 0G Compute to evaluate the safety vectors of a proposed action.

* Verifiable Cryptographic Signatures: Upon successful evaluation, the agent generates an EIP-712 cryptographic signature. Because the underlying inference was powered by 0G Compute and the reasoning logic is permanently anchored to 0G Storage, smart contracts can cryptographically trust these off-chain AI decisions. The entire computational lineage is secured, end-to-end, by 0G infrastructure.

Progress During the Hackathon

Week 1 — Integrating 0G Compute for Decentralized Agent Inference

The first major milestone was replacing centralized LLM calls with 0G Compute. We integrated the @0glabs/0g-serving-broker SDK directly into our LangGraph swarm, routing every critical agent decision through the 0G decentralized serving network. This included provisioning an account on-chain, dynamically selecting a 0G Compute provider, and building the inference request pipeline that the Researcher and Risk Guard agents depend on. The result: every intent parsed, every risk evaluation made, and every trade parameter derived by the swarm is now backed by a verifiable, decentralized AI call — not a centralized API.

Week 2 — Building Immutable Execution Proofs on 0G Storage

Once the inference layer was live, we tackled the data provenance problem. We built a full integration with the 0G Storage SDK @0gfoundation/0g-storage-ts-sdk), connecting to the Galileo Testnet indexer and storage nodes. The Reporter agent — the final stage in the swarm pipeline — was engineered to automatically package the complete execution metadata (intent, AI inference result, risk attestation, and outcome) and upload it to 0G Storage. Every completed session now returns a permanent rootHash from the 0G network, serving as an immutable, cryptographic "Proof-of-Execution." These root hashes are logged on the 0G Galileo chain explorer, publicly verifiable.

Week 3 — Autonomous Orchestration, Discord Trigger, and Live End-to-End Testing

The final phase was wiring together the full autonomous pipeline and running it live. We connected KeeperHub as the trigger layer so a Discord message autonomously initiates the full swarm cycle. The end-to-end flow now runs without any human intervention:

1. A user sends a natural language command in Discord

2. The swarm triggers, and the Researcher agent fires a 0G Compute inference call

3. The Risk Guard agent evaluates the output and generates a cryptographic attestation

4. The Reporter agent uploads the complete session log to 0G Storage and returns the Galileo root hash as on-chain proof

5. The Discord channel receives the final report with the 0G Storage scan link and root hash — publicly auditable by anyone

We confirmed multiple successful end-to-end executions live during the hackathon, with 0G Galileo explorer links and root hashes proving every agent decision made during the session is permanently stored and verifiable on the 0G network.

Still not raised any funds for this project, but will, for sure, in the future