Managing Prompts Without the Chaos: Simplifying AI Agent Development with Dakora

Mihai Lucian Androne — Sat, 04 Oct 2025 21:00:21 +0000

If you’ve built AI agents, you know the pain.

Prompts start as small strings. Then the project grows and now they’re scattered across files, duplicated, edited in production, and breaking things at runtime.

I’ve been there, and even had PMs ask to “just tweak a line” in a prompt. Without a clear system, every change meant touching code and redeploying.

So we built Dakora, a lightweight, type-safe, file-based prompt management tool, with a built-in playground that keeps things clean, predictable, and easy to iterate on.

You can try it right now via the online playground, no install needed: https://playground.dakora.io/

Real-World Use Case: Multi-Agent Orchestration with Microsoft Agent Framework

In this post, I’ll walk you through how I integrated Dakora into the newly released Microsoft Agent Framework by starting with a multi-agent orchestration scenario.

To make this work, I needed multiple distinct prompt templates:

One for the router (to analyze user intent and route requests)
Several for specialized agents, each with its own role and behavior (coder, researcher, writer, summarizer)
Additional templates for task flows like summarization or writing from research

With Dakora, all these templates lived neatly inside a /prompts directory, each defined in YAML, with typed inputs and descriptions. No more cluttered Python strings or manual string formatting.

Whenever I needed to tweak how the writer agent behaves or adjust the routing logic, I just edited the template in Dakora Playground and saw the change instantly, no code redeploy required.

You can explore the full example on GitHub, with step-by-step instructions for running the multi-agent demo.

The Dakora Playground: Edit, Test, and Iterate Fast

One of the best parts of working with Dakora is the Playground, a built-in web UI that lets you manage and test your templates interactively.
You can launch it locally with:

dakora playground

Here is how it looks in action for the multi-agent example:

With the Playground, you can:

🧭 Browse all templates in your project
✏️ Edit them with full syntax highlighting
🧪 Test different inputs instantly
💾 Save changes without touching code
🔄 Hot reload updates straight into your running agents

Try It Out

You can explore everything — including setup scripts, sample templates, and demos in the official GitHub repo

While we are at it, please ⭐ star the repo as it helps others discover Dakora and keeps the open-source momentum going.

If you want to dive deeper and add Dakora to your own project:

pip install dakora
dakora init
dakora playground

I built these examples because I needed a better way to manage prompts and Dakora turned out to be that missing piece.

If this resonates with you, check out the repo, run the examples, and let me know what you think.
Your feedback helps shape what comes next.

If your prompts still live as strings in your code… it’s time to level up! Stop fighting “prompt hell” and start managing them like real assets. Check out this practical guide on externalizing and editing prompts the smart way 👇 #ai #promptengineering

Mihai Lucian Androne — Thu, 02 Oct 2025 14:40:44 +0000

Stop Hardcoding Prompts: A Practical Workflow for AI Teams

Bogdan Pistol ・ Oct 2

#ai #llm #opensource #productivity

DEV Community: Mihai Lucian Androne

Totally agree!

Cut Your LLM Costs by ~30% With Prompt Optimization (What Actually Works in Production)