There are 304 workflow templates on KeeperHub, plus everything the community has published to the Hub. If somebody has already solved the thing you are about to automate, start from theirs. Your agent can take one and adapt it rather than writing from scratch.

A smarter model is not a safer one. More capability means more available choices, and more ways to get a transaction wrong at the moment it costs the most. That is why the layer underneath the agent has to be deterministic rather than clever.

The stablecoin rails for agents already exist. x402, MPP, settlement per call. What is still missing is a guarantee that the payment an agent authored is the payment that lands. Paying was always going to be the easy half.

Nobody notices a multisig ownership change until they need the multisig. KeeperHub reads Safe owners, threshold, modules and pending transactions natively, so a workflow can check the set you expect and page someone the moment it stops matching.

Idle stablecoins in a treasury are a decision nobody got round to making. Sky is 46 typed actions on KeeperHub. Your agent can author a workflow that routes a balance into sUSDS on a schedule or a threshold, inside a transfer cap the model cannot raise.

Add the KeeperHub MCP server to Claude, ElizaOS, CrewAI or LangChain. Describe what you want to happen onchain. Your agent authors the workflow. You review it and dry run it. KeeperHub then executes exactly that, on schedule. docs.keeperhub.com/agent/mcp…

An agent can call a tokenised equity contract at 2am. The market it tracks is shut. KeeperHub has a market status node for exactly that: halts, pause flags, stale quotes, feeds past their heartbeat and pending corporate actions, each named before a workflow acts on a price.

Correct code executing on wrong input is now a bigger category of loss than contract bugs. Staleness checks, a sanity band on price movement, a second source, a dry run before submission. None of that is research. It is work nobody schedules.

Most recurring payouts are still a script and a hot key, and nobody can reconstruct a specific run six months later. Your agent authors the run once, a human signs off before value moves, and every execution leaves a trigger, a transaction, a gas cost and a timestamp.

A health factor does not warn you before it crosses. It crosses, usually at 3am. Your agent writes the guard once: watch the position, and if the health factor sits under your threshold for three consecutive runs, repay. Reviewed by you, executed by KeeperHub.

For anyone who found us through the hackathon: the free plan is not a trial. A non-custodial wallet is provisioned when you sign up, your first run is gas sponsored, and there is a monthly execution allowance that does not expire into a paywall.

Agents decide, KeeperHub executes. Your agent authors the workflow once. You review it and dry run it without touching the chain. Then KeeperHub executes that exact artifact on schedule, inside policies the model cannot change, with a receipt for every run.

The bounty track asked for a mergeable KeeperHub feature rather than a standalone project. A LayerZero node and a state threshold trigger came in as pull requests. Anything that merges becomes a feature every other user gets, long after the event is over.

The brief was to integrate KeeperHub into a real project and land a real transaction. First went to Almanak strategies executed by KeeperHub, then Lucid Keeperhub, then bursar. The bounty went to a LayerZero node and a state threshold trigger, both as pull requests.

A channel with forty people in it is not an on-call rota. Nobody in it is responsible, and nothing escalates. KeeperHub now speaks @pagerduty natively, so a workflow pages whoever is on duty, and a condition that lasts all night stays one incident. keeperhub.com/blog/pagerduty…

Builders run your judging criteria through their own agents before they write any code. So we wrote the integration requirement into the criteria themselves. 289 builders, 140 submissions, and far fewer projects nobody could explain at demo time.

Hackathon ended with 140 submissions, 250+ registrants, 60+ PRs and 1000s of KeeperHub executions. Happy Friday and🍻to all our builders!

That's a wrap. Another great hackathon in the books, thanks again to all the builders, you've outdone yourself once again. Stay tuned for the next one ;)