Inspiration

I have been using multiple tools to stay organised, and I eventually give up. Some sort of documentation tool from cloud, to local first to only local, working with AI tools plan and design solutions to problems, some sort of time tracking, ticket systems, etc.

I coded and scripted ways to make them work together, it works for a while, then something happens, and I have to invest time on fixing it or changing the tool, then going back to it, you get it.

I started building HeyEddi.com as an agency, and then switched to a tool that would run the agency for me, at least the busy parts, like documentation, time tracking, tickets, prioritising emails, etc.

What I want is a team of AI coworkers orchestrated by an AI PM, that can work with Human coworkers too. That is the goal.

What it does

Its a Workspace administrator, a tool for small professional services teams or individuals, like Solo Founders Gemini acts as the operations layer on top of a unified workspace:

  • A vault where your AI Eddi learns about you, and stores your goals, skills, what you want to do, etc. This is private for your eyes only.
    • A workflow that goes from Strategy to Execution, AI Eddi can take a goal and scaffold up projects, new workspaces and help you strategyze with tools like Lean Canvas, Test Cards, Wardley Maps, and execute with a framwork (I Like Shape Up) and Kanban Board
    • AI Eddi will watch your workspaces and keep them updated as you work on them, think Confluence and Jira on Autopilot

How we built it

It uses a previos API that I was working on for local coding as the infrastructure. This API conencted to Zed's ACP and is OpenAI compatible too, so it can be used in Zed text editor or with any other CLI tool that can connect to a custom AI API. It runs Gemini Flash and lite, can use Pro too but I try to stay away form it due to pricing.

Then on top of it I started building a Firebase Service, with Firestore, Auth, Cloud Functions and Storage for this new HeyEddi Hub tool. Firebase is a great framework for productivity due to its ready to go pieces, and features. I use a lot of document watchers to update the UI and a document based DB really fits this project.

In General this is the idea:

  • AI: Gemini API (Vertex) via google-genai, orchestration, AI PM proposals, vision for receipts
  • API: Cloud Run — FastAPI gateway with multi-tenant auth
  • Data: Firestore — workspaces, vault, nodes, tasks, conversations, usage
  • App: Firebase Auth + Hosting — Nuxt 4 static site

Challenges we ran into

I think the most challenging part is how to provide a useful service, I did not want just a another chat tool. So thinking about what things we can do with out full approval but reviewed, more like a Human to Human collaboration looks like, so I landed in a more Human framework when we have an Orchestrate that talks to other Agents ad Humans via the Documentation and Tickets, instead of calling all AI agents directly, this way we can leave a paper trail, give feedback, disagree, etc. Just like any normal workday.

In this Hub, the AI Agents are not your tools, they are your peers.

Another issue, since this is multi-tenant, is privacy, so I came up with the Vault concept, to have a private place for every user, and workspaces where users can invite others to work on. Firestore Rules are great for this, AI Eddi can only see the workspaces you have access to, and determine what you are working on, which project, which workspace, which ticket. So you can use the tool from the chat, or from the UI, the idea is to make it simple and secure.

Accomplishments that we're proud of

  • Gemini in production, not in a slide deck, every orchestrated request hits the Gemini API. Agent execution, usage tracking, and proposal merges run on Cloud Run and Firestore today.
  • Vault + Context Bridge, a privacy architecture we’d trust with real financial and career data, wired into live companion chat.
  • AI PM approve/merge loop. Journal to structured proposal to owner approval to production write. AI-native operations with a human gate.
  • One graph, many framework views, IU realized early that I wanted to use multiple ways to show data, so I ended up with a universal schema that suports front end Wardley and Lean Canvas, and the KanBan too

What we learned

AI-native means operations, not features., not a chatbot and the edge of the screen. People need a single tool for all aspects of life, this is a very common productivity tip, but to do that you always end up with another tool and try to sync work to it. I want to bridge that, a single tool to bring all of that automatically, starting with people that don't necessarily have another tool already, like Solo Founders, small teams. Eventually I would sync tools for Enterprise like Jira, Line, or whatever else is feasible

What's next for HeyEddi Hub

Well, I started showing the first screens to people, trying to see if they get what the tool is. After I have enough working features I want to get some Paid pilots, and prioritise the next features based on their feedback.

Built With

Share this project:

Updates