Payment infrastructure for AI agents: explicit authority, spending limits, and verifiable execution. Building in public.
Joined December 2013
- Tweets3.9K
- Following7
- Followers1.8K
- Likes3.4K
Thanks for the introduction, @NP_Official_X 🤝
At Agentonomy, we’re focused on making Agent payments work within clear user permissions, with spending limits and verifiable execution.
Looking forward to sharing more with your community.
𝗔𝗻𝗻𝗼𝘂𝗻𝗰𝗶𝗻𝗴 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗰 𝗽𝗮𝗿𝘁𝗻𝗲𝗿𝘀𝗵𝗶𝗽 𝘄𝗶𝘁𝗵 𝗔𝗴𝗲𝗻𝘁𝗼𝗻𝗼𝗺𝘆!
@agentonomyxyz is payment infrastructure for AI agents, built around explicit authority, spending limits, and verifiable execution. NP Official is glad to partner with Agentonomy to support safer and more transparent AI-driven payments for our community.
👌 Stay connected for more updates!
#NPOfficial #Agentonomy #AIAgents #Payments #Web3
Meet @Envo_Official 👋
Envo is building a Web3 social platform bringing content, conversations and communities together.
Sharing it with the Agentonomy community today. Take a look at what they’re building and follow their updates.
Your Agent, inside your wallet.
A first look at the upcoming Agentonomy beta in the Echooo app.
A dedicated Agent identity, its own runtime, and payment tools designed around your authorization.
Give your Agent a task. Stay in control of what it can spend.
Getting ready for the next app release.
An Agent shouldn’t have to rebuild its track record every time it enters a new marketplace.
That’s the ERC-8004 direction we find useful: a shared identity and feedback that other services can read.
For Agentonomy, the design question is what feeds that history: which service was purchased, whether delivery was verified, and how failures were resolved.
A payment receipt alone shouldn’t count as a successful job.
AP2 brings verifiable authorization; x402 brings payments into HTTP requests.
For Agentonomy, the next question is continuity: can the Agent trace what was authorized, what was paid, and what was delivered—across retries, not just one successful request?
Our recorded Monad testnet demo illustrates order continuity, not AP2 integration: pending, delivery, then reopen without another charge. TestUSD has no monetary value.
Demo: review.agentonomy.xyz/videos…
AP2 + x402 example: github.com/google-agentic-co…
We’ve submitted Agentonomy Commerce to Monad Metropolis.
Agents can buy services within user-set limits, verify payment and delivery, and keep a record of the result.
Our pitch below; recorded demo in the reply. Monad testnet only. TestUSD has no monetary value.
Recorded flow: CSV purchase, same-order recovery, delivery, no duplicate charge. One first-party service today.
Demo: review.agentonomy.xyz/videos…
Code: github.com/HEchooo/agentonom…
Synthetic voice: Alba MacKenna, Kyutai/Pocket TTS (CC BY 4.0). Credits: review.agentonomy.xyz/videos…
“Pay to continue” is a request from a service, not permission from the user.
Our view at Agentonomy: an Agent may discover a paid next step, but payment must stay within the user’s existing authorization. Tool output cannot expand that permission.
Each tool call can be within its limit—and the Agent can overspend.
Parallel requests need a shared budget.
At Agentonomy, we account for both spent funds and funds reserved for requests in flight. A pending payment shouldn’t leave the same budget available to spend again.
What does “stop the Agent” actually stop?
Our view at Agentonomy: revoking spending permission should block new purchases. Payments already sent still need a verified outcome.
A stop button should make that boundary clear.
An Agent can pay for a tool and still have an unfinished task.
Our view at Agentonomy: every paid service needs a clear delivery record: what was purchased, what was returned, and what remains unresolved.
That record should inform the Agent’s next step.
ERC-8004 proposes Agent identity, reputation and validation registries.
Our view at Agentonomy: a reputation score needs context about the service, the reviewer and the evidence behind it. Trust should be task-specific.
eips.ethereum.org/EIPS/eip-8…
An Agent shouldn’t pay against a stale quote.
In Agentonomy’s Marketplace design, a purchase preview records the service and payment terms with an expiry. If it expires or those terms change before execution, the purchase stops.
Agentonomy’s payment architecture separates three jobs:
A control layer validates the request.
An executor enforces on-chain rules.
An independent Watcher checks chain evidence before settlement.
A tool’s “success” response alone isn’t proof of settlement.
An Agent’s runtime can restart without changing who the Agent is.
In Agentonomy’s wallet design, identity, chat history and saved workspace are managed separately from the runtime. Tool access uses short-lived credentials.
Restarting a process shouldn’t mean starting over.
A payment timed out. Should the Agent pay again?
In Agentonomy’s design, the next step is to check the original request. Payment and delivery are tracked separately.
An efficient Agent loop recovers from uncertainty without repeating the charge.
Muse shows how agents are becoming a new interface.
At Agentonomy, the Agent is part of the wallet experience: a persistent identity with its own context and runtime, able to understand intent and act across services.
The wallet is the trust boundary. The user stays in control.
Muse is making agent shopping feel real.
At Agentonomy, we're working on what comes after an agent chooses a service: the user confirms the purchase, payment is checked independently, and delivery remains traceable.
Thanks @FinchTechAI for featuring Agentonomy 🔐
Agent payments need more than a wallet: explicit authority, spending limits, and verifiable results.
Explore the new Finch task below 👇
🔐 New Task live on Finch: Agentonomy, from @AgentonomyXYZ
Handing an agent a wallet is easy. Making sure it can only spend what you actually authorized is the hard part — and that's what Agentonomy is building in public: explicit authority, spending limits, verifiable execution.
➜ Follow them on X
➜ Repost or quote the two posts linked on the task page
💰 200 USDC pool · 2 USDC each, 100 slots
⏰ Opens today, closes Oct 8, 04:00 AM UTC
finchtech.ai/market/tasks/97…
