💡 Inspiration
Every team building an AI agent repeats the same setup: tools, models, instructions, monitoring. In finance, a wrong number is worse than no answer, and every dollar of spend needs an owner. We wanted a platform where people only decide what the agent should do, for developers and non-developers alike.
⚙️ What it does
- 🪄 Describe a goal, get a draft. A setup assistant fills in the form. A person reviews and clicks deploy.
- 👥 Two ways to build. A web form for non-developers, a simple YAML file for developers. Both make the same agent.
- 🔌 Connect tools with MCP, limited to connectors the admin approves.
- ✅ Test before shipping. Exact checks catch wrong numbers and missing citations. A second model grades the rest.
- 💰 Control cost. Every model call is measured. Admins get a warning at 80% of budget, and at 100% the agent switches to a cheaper model so work still finishes.
- ♻️ Share and reuse. Anyone can copy an agent from the catalog. Only the owner can edit it.
Built With
- claude
- docker
- fastapi
- github
- langfuse
- langgraph
- mcp
- meta-model-api
- next.js
- postgresql
- python
- typescript
- vultr
- yaml
Log in or sign up for Devpost to join the conversation.