Defi Preacher

Hey devs

As devs we automate everything, except the thing we do DAILY:
writing commit messages.

So I fixed it.

Introducing Commitra⚡️
Your AI commit assistant that writes clean commit messages and auto commits for you in 1 sec.

No more thinking.
No more “final-final-FINAL commit”.

Just:
git add .
commitra commit

Groq is the default provider (free & blazing fast).
Setup in 10s → commitra config set GROQ_API_KEY=your_key

Also, you can generate readme, api-docs, workflow diagram, and folder tree for your project with it

npm: npmjs.com/package/commit…

Somewhere along the way, giving AI agents more tools became the answer to almost everything.

But what happens when you have all those tools, and getting them to work together is still a job of its own?

Say you run a community on Discord, track issues on GitHub, and talk to your team in Slack.

You want something simple: "whenever someone reports a problem, open an issue, tell the team, and keep the person updated until it's fixed"

That's one sentence.

Setting it up is another story: which tools to use, connecting the services, finding channel IDs, permissions, handling failures, and making sure it still works tomorrow.

At some point you're spending more time managing the automation than doing the work you wanted to automate.

So for the @NousResearch Hermes Agent Plugin Hackathon, I built Capalis.

Capalis turns what you ask your agent to do into a real, reusable workflow that keeps running without you coordinating every step.

It works inside Hermes Agent. You describe the job in ordinary language. Capalis finds your channels and repository by name, works out which of your connected tools can do each step, and recommends an approach. You can take the recommendation or compare the options yourself.

Then it puts the steps together into what I call a Recipe: how the job should be handled, where each action belongs, and what happens next.

Before anything real happens, it practises the whole job once without sending anything. You see exactly what it will do and where, and you approve that exact version from your phone on Telegram.

Once approved, the Recipe runs whenever the right event happens, or on a schedule. You don't repeat the instruction. It can wait for later events, pick up after an interruption, and remember what it has already done.

It reads back each result from where it landed, so an uncertain response never turns into doing the same thing twice. And if a tool changes, a permission stops working, or something the Recipe depends on is no longer valid, it stops that workflow and asks you to review an updated version.

I wanted people to have that control without needing to understand the engineering underneath.

For the demo, everything is real: Discord, GitHub, Slack and Telegram accounts.

A bug reported in Discord becomes a GitHub issue. The team gets a heads-up in Slack. The reporter gets a reply, and when the issue is closed, Capalis follows up in the same conversation. I approve it from Telegram, and I can pause it from there too.

I also ran a different job: checking a website on a schedule and telling the team in Slack when it's down.

Capalis isn't meant to be another tool for one problem. It's meant to help Hermes make better use of the capabilities it already has, combine them into useful jobs, and keep those jobs dependable over time.

Under the hood: a controlled runner, a durable record of every step, approval and safety checks, and 305 automated tests, including ones that deliberately break things mid-run. The repo also has the evidence from the real runs.

The most exciting part of building with AI agents, for me, is when we stop thinking about the individual tasks they can do and start thinking about the whole jobs they can take responsibility for.

You should be able to describe what you want done, choose how it's done, approve it, and get back to what actually needs your attention.

Here's Capalis demo in action 👇

0:00 The problem
0:10 Describe the job, and Capalis builds it
0:21 Approve it from Telegram
0:31 Discord, GitHub and Slack working together
0:49 It follows up when the issue is closed
1:00 You stay in charge: pause it, and it stops itself when something changes
1:12 The proof, and how to install

Install: hermes plugins install Femtech-web/Capalis#plugin
Repo: github.com/Femtech-web/Ca…
Catalog PR: github.com/NousResearch/h…

One instruction. The right tools. A workflow that keeps working.

gm!

There’s quite a lot to look forward to in the @zama ecosystem right now.

Season 5 of the Developer Program is live, and I particularly love that the focus is on building games where confidentiality actually makes a difference to the gameplay.

It’s a good opportunity for builders to experiment with encrypted game states, hidden information, and things that would be difficult to achieve on a fully transparent blockchain.

And if you’re just getting started with FHE, don’t let the technology intimidate you. Start small, understand the fundamentals, and build from there.

I’ve been sharing what I’ve learned through my Inside Zama series, and there’s still plenty more to cover.

Looking forward to seeing what everyone builds this season. I

Have a great day! 💛

Zama cooked with FHE!

The world doesn’t have to know your figures onchain

Send yours too at app.zama.org

“Can we just add one more thing?”

If you've worked with clients as a freelancer, you probably know how quickly that one sentence can turn into extra features, more revisions, and sometimes, work you never agreed to do in the first place

I've experienced this myself as a software engineer, and honestly, it's one of those things I've always wanted a better way to handle.

Then there's the other side of freelancing. Keeping track of what was agreed, what's been delivered, who's paid, what's still outstanding, and all those little promises scattered across conversations.

When I came across the Tatum × Walrus Total Recall hackathon, I saw an opportunity to finally build something around a problem I understood firsthand.

And the more I looked into it, the more I realized I wasn't alone. Other freelancers were dealing with similar issues: clients quietly expanding project scopes, forgotten invoices, delayed payments, awkward follow-ups, and disagreements over what was originally promised.

So I built Parius

It's a shared workspace where freelancers and clients can communicate, agree on project terms, review work, manage changes, and keep track of payments without losing the context of their conversations.

When a client requests something outside the agreed scope, Parius can help identify it and suggest how to respond. But the final decision always stays with the people involved.

And here's the part I particularly love.

Parius actually remembers.

Using Walrus Memory, it can learn from confirmed decisions and past client experiences, then bring those lessons into future projects.

So instead of repeatedly dealing with the same problems from scratch, you have an assistant that remembers what you've learned along the way.

I also integrated Tatum to verify USDC payment transactions on @SuiNetwork, helping connect actual payments to the work and invoices they belong to.

The first version is live now. It's still early, and there's plenty I want to improve, but I'm glad I finally turned a problem I've personally dealt with into something useful.

It’s very easy to use too and you have a doc on the platform incase you need anything

Give Parius a try: parius.femdev.site

Built for the @tatum_io × @WalrusProtocol Total Recall hackathon.

blog: dev.to/adeyemi_femi_f…
code: github.com/Femtech-web/Pa…