Agents Production (Playground)

TypeScript scaffolding to build and evaluate AI agents with a minimal dashboard and OpenAI integration.

A production-leaning playground for running AI agent experiments. It provides a clean TypeScript entrypoint, a structured place for agent code (tools, routers, memory), a lightweight dashboard scaffold, and an `evals/` workspace for reproducible evaluation runs. Designed to iterate fast on prompts, tools, and behaviors while keeping the project layout predictable.

Tech Stack

TypeScriptNode.js (20+)BunOpenAI API

Features

TypeScript app entrypoint (`index.ts`) to launch agent logic
`src/` for agent code, tools, memory, and routers
`evals/` folder for prompts, fixtures, and run logs (reproducible evals)
`dashboard/` scaffold for a local UI you can evolve
Node.js or Bun runtime; simple npm/bun scripts
OpenAI API integration via environment variable

Challenges

Designing a minimal yet scalable structure for multiple agent workflows
Keeping evaluations repeatable and comparable across runs
Balancing fast iteration with type-safety and clear boundaries

Learnings

Set up a repeatable evaluation harness for agent behaviors
Standardized folders for tools/memory to speed up agent iteration
Environment-based wiring for provider keys and runtime settings

Feedback

For feedback or suggestions, contact me at: raghavkhandelwal847@gmail.com

Link

Github