Accord

• 

  Marketplace Page

• 

  Agent Page

• 

  My agents Page

• 

  Home page

• 

  Analytics Page

Inspiration

The project story is simple, we need a central place to create agents, a place where users can create the agent they want, let it run and test itself and continually improve while learning from real use by the user or other users.

What it does

What accord does is simple, it gives the user the ability to prompt the agent they want, they simply need to say, "Hey create an agent that checks my emails and answers them for me" all while selecting the gmail tool and voila, that's it, a live email agent that has recursively checked itself for errors, and consistently checks itself for errors based on tracing data in real and test use cases. It also gives users the ability to add their agent to a marketplace in order to earn cash through their agent use if necessary.

How I built it

Well I did brainstorm a lot but to be honest it was really AO on my windows machine that did lot of the building, while I did manage it over about 2 agent harnesses (Opencode and Claude Code), I do have to say I found myself relying on the app more, I did do some coding here and there but this was after my credits had run out. It uses e2b to actually create a separate system to test and run the agent in an isolated environment.

Challenges I ran into

AO on windows is buggy, I had to fight the software for an hour or so simply to setup the model harnesses, not to mention there are a lot of issues the desktop app has, like session issues, model picking issues and terminal issues, some of these may be edge case, but while the app is impressive I can't help but imagine it being a little "beta".

Accomplishments that I'm proud of

I am super proud of the speed of output, built the homepage and a lot of other pages as well as important backend code in a span of hours, this is very impressive even by my coding standards.

What I learned

Learned about AO more, and using agent orchestrators, this one is really really cool, I learned a bit more about design decisions, looking at my code will give you a lot of the design decisions I chose to make this a success, there is soo much poured into how the backend carries out this task.

What's next for Accord

Building it properly, there is no doubt that the platform has real merit, we see startups that their whole thing is actually a maketplace for scrapers of all kinds, why not bring that into the real AI usecase: Agents. That way people/professionals can actually use an agent they can get trace level reporting on and see the agent improve in real time.

Built With

• agent-orchestrator
• ao
• nuxt
• python