Most trading agents still need full wallet access.
Agon gives them a controlled trading budget instead, with limits the agent cannot override.
You can describe a strategy, let the agent execute it, and keep control over how much it can actually do.
We’re building it for the Colosseum hackathon and opening up the waitlist now.
Waitlist: waitlist.getagon.tech
X: x.com/agonpilot
We're building a project for the Colosseum hackathon around giving AI agents restricted trading permissions instead of handing them a full wallet.
One failure case we're trying to reason through:
an agent submits a swap, the RPC request times out, and it doesn't know whether the transaction actually landed.
If it rebuilds and signs another transaction, both could potentially execute. A spend cap limits the damage, but doesn't distinguish a retry from a genuinely new trade.
What we're considering:
Would you enforce any notion of trade/intent idempotency on-chain, or is transaction-level deduplication + reconciliation at the executor enough?
Also curious about the uglier cases: RPC disagreement, blockhash expiry around a retry, and multiple actions being in flight at once.
Trying to break the design before we ship the hackathon build, so would appreciate any failure cases we're missing.
context first: this is a hackathon project, and i'm not selling anything. i've been posting in a few communities for a while to get real feedback. you can check my profile. every round, people told me what was unclear or missing, so i went back and reworked it. the latest update adds a TUI (terminal UI) so you can use it straight from the terminal.
the problem: AI agents can now trade and touch wallets. but letting an agent make decisions and giving it unrestricted authority over your funds are two different things.
the approach: Agon is an agent-agnostic control layer, not another trading agent. you pick any agent, and separately control how much authority it gets. the short version is give your agent a budget, not your keys. we're also looking at a trader's own history to suggest the limits they already tend to follow, instead of generic rules.
what i still want honest feedback on:
We're building Agon for the current Solana Crypto World's Fair, and I'm looking for people to poke holes in the idea rather than just tell us it sounds cool.
The problem we're thinking about is this:
AI agents are getting increasingly capable of trading and interacting with wallets, but giving an agent the ability to make decisions and giving it unrestricted authority over your funds feel like two very different things.
Agon is our attempt to separate those.
Instead of being another trading agent, Agon is being built as an agent-agnostic control layer. You can use the agent you want, while separately controlling how much authority it gets.
We're also looking at a trader's own historical behaviour to help surface the rules and limits they already tend to follow, rather than inventing generic rules for everyone.
We're starting with a read-only beta. Users connect the Solana wallet they actually trade from and sign a free message to prove ownership. No transaction and no access to funds.
The questions I'd genuinely love feedback on are:
Does separating the agent from its financial authority feel like a real problem worth solving?
Would being able to switch agents while keeping the same guardrails matter to you?
What would you need to see before trusting an AI agent with any amount of real funds?
Is there something obvious we're missing about this model?
Current beta/waitlist: waitlist.getagon.tech
Happy to hear the harsh version. We're still shaping the product and that's much more useful to us right now.