Agenta
The open-source workspace for building, running, and managing specialized AI agents.
Project Gallery
Project Overview
Agenta is a comprehensive, open-source workspace designed for building specialized AI agents that automate and augment complex workflows. The platform allows users to build agents interactively via chat, connect them to external applications, and improve them through continuous feedback. It features robust versioning, execution tracing, and asynchronous background queues, allowing agents to run on schedules or trigger via real-time events. By supporting almost any LLM and harness (like Claude Code and Pi), Agenta provides a flexible, collaborative environment for teams to build reliable, scalable AI integrations without vendor lock-in.
Tech Stack
Next.js
TypeScript
Python
FastAPI
PostgreSQL
Redis
Tailwind CSS
JotaiKey Features
Actively contributed to the core open-source repository by developing new features, resolving issues, and collaborating with the maintainer team.
Contributed to developing a collaborative open-source workspace supporting team access controls and version history for AI agents.
Engineered features for a scalable backend using FastAPI and PostgreSQL, utilizing Redis and Taskiq for asynchronous background agent execution.
Implemented robust observability and tracing to inspect model calls, track token usage, and estimate costs per agent run.
Built a flexible integration layer supporting MCP (Model Context Protocol) servers and over 1,000 apps via Composio.
Created a modern, highly reactive Next.js SPA utilizing Jotai for complex state management and Tailwind CSS for responsive design.
Case Study
The core challenge was providing a unified environment where users could interactively chat with agents while also deploying them for non-blocking, asynchronous background tasks. This was solved by separating the interactive Next.js frontend from a scalable FastAPI backend that offloads scheduled and event-driven tasks to background worker queues. Additionally, ensuring reliable execution across a massive variety of external LLMs required building an abstracted routing and tracing layer that tracks token usage and intercepts tool calls seamlessly.
Problem Space
Challenges
Actively contributed to the core open-source repository by developing new features, resolving issues, and collaborating with the maintainer team.
Contributed to developing a collaborative open-source workspace supporting team access controls and version history for AI agents.
Engineered features for a scalable backend using FastAPI and PostgreSQL, utilizing Redis and Taskiq for asynchronous background agent execution.
Our Approach
Solutions
Implemented robust observability and tracing to inspect model calls, track token usage, and estimate costs per agent run.
Built a flexible integration layer supporting MCP (Model Context Protocol) servers and over 1,000 apps via Composio.
Created a modern, highly reactive Next.js SPA utilizing Jotai for complex state management and Tailwind CSS for responsive design.
Explore More
Like what you see?
Check out more of my work or reach out to collaborate on something exciting.