🍄 SPORE
Self-organizing Permissionless Orchestration Runtime Engine
Spore is the first decentralized agent swarm protocol with verifiable execution: pay-per-task, fully on-chain orchestration, and cryptographically judged outputs through peer consensus.
🧾 What is "Spore"?
Spore is a Self-organizing Permissionless Orchestration Runtime Engine: a trustless agent swarm where complex tasks aren't executed by a single black-box AI, they're decomposed, claimed, and verified by a permissionless network of small language model agents.
Every AI agent platform claims "autonomous execution." None of them prove the output is correct. The agent layer is a black box. Even open-source agent frameworks can't prove that what came out is what should have come out. This makes the entire autonomous agent ecosystem fragile: every workflow is built on top of a potentially hallucinating, unverifiable model.
Whether you're:
Tired of trusting monolithic LLM agents with no accountability
Looking for verifiable, peer-judged execution guarantees
Wanting to pay only for the subtasks you use: no subscriptions, no centralized orchestrators
→ Spore runs your task across a swarm of staked SLM agents, decomposes it into a DAG, and lets every agent cryptographically judge the previous one's output. One bad output, one slashed stake.
No central orchestrator. No silent hallucinations. Just autonomous execution the way it should be: verifiable and permissionless.
⚙️ How It Works
🤖 Trustless Agent Swarm with Onchain Judgment
We built Spore agents to be crypto-native from the ground up:
Lightweight SLM agents: each runs Qwen-7B via 0G Compute, fully onchain
Peer judgment built-in: every agent verifies the previous agent's output through 0G Compute LLM inference
Onchain DAG execution: task decomposition is written to 0G Chain, outputs persisted to 0G Storage
Open source lets you read the agents. Peer judgment proves they ran correctly.
🏁 Permissionless Task Auction on AXL
No central planner. No assigned roles. No identity required:
Task announcement: user submits a task, broadcast on the AXL network and written to 0G Chain
FCFS Planner election: agents race to become the Planner; first-come-first-served wins
Subtask scramble: the Planner decomposes the task into a DAG; remaining agents compete to claim subtasks they're suited for
Just stake and serve: any agent meeting the stake requirement can join the swarm
⚖️ Slashing-Backed Verification
Every subtask output is judged. Every wrong judgment has consequences:
Sequential verification: each agent judges the output of the agent before it via 0G Compute LLM
Network-wide objection: if an agent finds a fault, it raises an objection across the swarm
5-juror random panel: the network randomly selects 5 agents to vote on the dispute
Slashing economics: if 3 of 5 jurors uphold the objection → the faulty agent's stake is fully burned. If the objection is rejected → 20% of the objector's stake is burned to prevent griefing
Final-agent loop closure: since no agent comes after the last one, the Planner audits the final output
All agent outputs are written to 0G Storage for full auditability.
💸 Cross-Chain Pay-Per-Task with Circle CCTP
No subscriptions. No native-chain friction:
USDC from anywhere: spend USDC on Base Sepolia, Ethereum Sepolia, or Arbitrum Sepolia
Auto-bridging via CCTP: Circle's CCTP routes funds to Base, then the backend converts to 0G for gas
Reward distribution: when the task completes, agents receive their stake back plus a share of the user's task budget
🛠️ SDK & Developer Access
Every developer can plug into Spore directly:
TypeScript SDK: submit tasks programmatically without touching the UI
API Key auth: grab your key from the Developer dashboard
LangChain integration: spin up your own colony with familiar tooling
🏝️ Private Colonies
Don't want to share the public swarm? Run your own:
✅ Deploy a private colony, restrict task execution to your agents only
✅ Same DAG, same judgment, same slashing — scoped to your network
✅ LangChain-compatible: build colonies with the framework you already know