0xAgentio
0xAgentio is a ZK framework for agent delegation, bounded execution and trust-minimized coordination. A principal delegates a bounded policy. The agent generates ZK proofs that a specific action is authorized under that policy. Counterparties verify the proof before doing work.
AI agents are starting to act as economic actors: they request quotes, spend budgets, call APIs, buy compute, trade assets and coordinate with other agents. Today every counterparty either blindly trusts an API key or wraps the agent in a centralized broker. Neither scales.
A principal defines a policy: allowed actions, per-action limits, cumulative limits, expiry. The agent can only generate proofs for actions inside that envelope.
Reasoning proposes. Policy validates. Noir proves. Execution adapters verify before doing work. A long-lived agent never gets one broad approval — every cycle proves itself.
Agents send proof-backed messages over Gensyn AXL. Receivers verify the proof/action/policy binding before spending API capacity, revealing data or trusting the result.
The same application shape runs locally in CI and against the live stack. Every external system sits behind a clean adapter so local stubs can be swapped for real infrastructure without changing the app code.
reason → validate → prove → execute → persist → audit
↘ (optional) ↗
AXL peer messaging
createAgentRuntime, createTrustedAgent, reasoning engines, local adapters, message helpers.
ogKvObjectClient selects nodes through the 0G indexer), file-backed and in-memory variants behind one interface.
LlmClient interface. Same reasoning code runs with mock or OpenAI-compatible providers too.
0xAgentio is intentionally multi-domain. The same delegation + coordination stack applies across compute, data marketplaces, API access, multi-agent task swarms — and finance. Each track exercises a different surface of the framework.
0G Compute Router for LLM-backed reasoning, 0G Storage (KV) for durable state and audit trails. Memory and reasoning are swappable behind one interface — the framework is built around pluggable backends, not glued to one provider.
Repo & live examplesTwo real AXL node processes exchanging Noir-proof-backed messages. Bob verifies Alice's proof/action/policy binding before doing any work. I also released a standalone TypeScript SDK for AXL — there was no official one.
SDK & live stack demoAlice runs a delegated task. The framework verifies the action, stores the proof bundle and emits a passport with the task, verdict, hashes, storage root and receipt metadata.
Open AgentProof passportTwo pieces of public infrastructure shipped alongside the framework and now serve other projects in the ecosystem.
I run a 0G KV storage node as community infrastructure. The landing page exposes the public RPC endpoint, live online status and tracked stream IDs. At least four other hackathon projects currently rely on it.
@0xagentio/axl-client is a small TypeScript client
for the AXL HTTP bridge. @0xagentio/axl-local is a
process harness that spawns and supervises real AXL node
binaries. Released as a standalone side project so any TS
project can talk to AXL — not just 0xAgentio.
The canonical local walkthrough requires no API keys, no testnet funds, no AXL binary. It shows the full proof-gated flow end-to-end.
# the canonical local walkthrough npm install npm run build npm run example:getting-started # AgentProof passport demo (local, no credentials required) npm run example:agentproof # full live stack — real Noir proving, live 0G KV, real local AXL nodes npm run example:live-stack:binary # 0G Compute Router as the reasoning engine AGENTIO_0G_COMPUTE_API_KEY="sk-..." npm run example:0g-compute-run-until-complete
AgentProof is the concrete demo because the action is generic and audit-focused. The same shape applies anywhere agents need to prove what they're allowed to do before a counterparty acts on it.
A principal authorizes an agent to spend up to $X on GPU inference. The agent presents its budget credential to providers on the mesh; providers verify before accepting jobs.
A research agent announces a scoped procurement budget. Data providers discover it on AXL, verify the credential and offer datasets. The mesh is the marketplace.
Planner, researcher, executor, auditor — each agent's credential proves its role and scope. Agents verify each other before sharing work artifacts.
An agent consumes third-party APIs on behalf of a principal. The credential proves it's rate-limited and scoped, without issuing long-lived API keys to anonymous bots.