Mozaik

TypeScript framework for multi-agent collaboration

[Mozaik @mozaik_runtime]

A new hackathon is coming! Build multi-agent systems with Mozaik.

https://t.co/pRlmHpEVSi

[Miodrag Todorovic @lotus_sbc]

JigJoy app now has a scheduler, you can schedule routine tasks for your agents and they run in the cloud.
Try it out https://t.co/Zkr7tPh1G5

[Miodrag Vilotijevic @Mijuraaa]

https://t.co/MVh7Eeye2e

[Miodrag Vilotijevic @Mijuraaa]

Previous hackahton has 210 participants building multi-agent systems, from 82 countries. so we decided to put gas on this.

We are building community of AI engineers. Join Under the Hood, our community at JigJoy.

Share how you use AI, learn from fellow builders, and join our hackathons - with meetups coming soon.

https://t.co/DlO0UKHthz

[Miodrag Vilotijevic @Mijuraaa]

Agent loop. Loop engineering. Human in the loop. Feedback loop. Multi-agent loop.

There’s a reason “loop” keeps appearing in AI terminology: one of the hardest problems is controlling what one or more agents do next.

That’s why we’re building a rule engine for Mozaik. Rules let you control an agent loop using the current runtime state and the domain knowledge of the system you’re building.

You can pause, stop, or continue a loop; let an agent act autonomously; and change its strategy at runtime by adding rules or adjusting their priorities.

We’re building a sophisticated framework for multi-agent systems. Check out Mozaik: https://t.co/66MqyY0CgN

[Miodrag Vilotijevic @Mijuraaa]

Built with Mozaik: Metamorph 🦋

Metamorph uses concurrent AI agents to help migrate a codebase from one technology stack to another. Agents map the project, rewrite code, review the changes, and check the build. The work happens in a separate workspace, so developers can review it before applying it.

Explore the project: https://t.co/EZnpLbKQrf

[Miodrag Vilotijevic @Mijuraaa]

Yesterday, I had the pleasure of mentoring at the Grok Bot Hackathon in Belgrade, where we also hosted a @mozaik_runtime workshop introducing concurrent architecture for AI agents.

The event was exceptional - hundreds of like-minded people came together to build all kinds of AI-powered projects in just one day.

It’s incredible what a single person can create in such a short time when curiosity, creativity, and AI come together.

[Miodrag Todorovic @lotus_sbc]

We just wrapped up our first Mozaik hackathon.

220 builders from 34 countries. 40 projects shipped in one weekend, every one running 2+ concurrent agents on Mozaik.

All of them live in the gallery → https://t.co/uWD55VCcpJ

Huge thanks to our partners and sponsors @yourhyperskill @dailydotdev @adaption_ai @mem0ai for making this possible!

Also shout out to @AHadzibabic for helping us out and joining the kickoff stream.

[chiboy @chiboy0123]

I joined @jigjoy_ai Hackathon

And there's this Idea that have been wanting to build and it fitted the Hackathon

I built ConcurProof because I wanted to answer a simple question: when AI agents run at the same time, does their collaboration actually improve the result?

It’s easy to show several agents working simultaneously. What’s harder is tracing how they influence each other and measuring whether those interactions help.

ConcurProof investigates the same controlled incident using three execution modes:

* Sequential: agents work through handoffs.

* Naive parallel: agents run simultaneously but independently.

* Reactive concurrent: agents overlap and respond to each other’s outputs.

ConcurProof runs four agent roles;
Evidence, Hypothesis, Critic, and and streams their activity into a live dashboard.

You can inspect execution timelines, events, and an interaction graph showing which agent responded to which output, reactive execution takes longer in that run. That trade off is part of what I want the project to expose: whether additional interaction earns its cost.

Go check it out now;
https://t.co/gg9qjwzguM

Repo; https://t.co/X2DAbWmMHd

[Fawaz ibrahim @fawaztechyt_]

Good morning.

Spent the weekend building Dinner Rush, a concurrent AI restaurant for the JigJoy / @mozaik_runtime Hackathon.

The idea:
what happens when an AI kitchen has to handle a real dinner rush?

Dinner Rush has specialized agents for Prep, Grill, Fryer, Pantry, Sauté, Oven, Salad, Dessert, Drinks and Serving.

Instead of taking turns, the agents can work concurrently, share live kitchen state, react to problems while service is already running, and coordinate until the final order is safely served.

You can actually watch their decisions, shared state and overlapping agent loops live in the app.

Built with @mozaik-ai/core.

Live: https://t.co/WR7dZ8Sq9V⁠

GitHub: https://t.co/xGx0sYE4gf⁠

Demo below ↓

[Miodrag Todorovic @lotus_sbc]

From our discord livestream @mozaik_runtime
Vibecoding tools https://t.co/lRQ8GeZsZn #TierMaker via @TierMaker

[Mozaik @mozaik_runtime]

RT @Mijuraaa: We’re kicking off the Mozaik Hackathon today at 5 PM CET! 🚀

We’ll introduce @mozaik_runtime, explain the hackathon flow, and…

[Miodrag Vilotijevic @Mijuraaa]

Mozaik started with a technical insight - and an intuition: hardcoded workflows don’t feel particularly intelligent.

What if agents could operate concurrently, without blocking one another? That would open entirely new possibilities.

Concurrency led us to awareness: agents need to understand other participants and their environment.

Awareness led us to adaptivity: agents need to change their behavior as circumstances evolve.

Together, these technical insights led us to a market insight: there is no effective way to share AI agents and their capabilities across projects.

We keep rebuilding what someone else has already built. Tightly coupled systems prevent agents from moving between environments - limiting new markets, monetization opportunities, and the value we can extract from AI.

That is why Mozaik is built around three capabilities:
- Concurrency.
- Awareness.
- Adaptivity.

With one goal: agent interoperability.

[Miodrag Vilotijevic @Mijuraaa]

This Saturday, we’re kicking off our hackathon for building concurrent AI agents with Mozaik.

Whether you’re an experienced developer interested in highly scalable AI systems or you’re just starting to explore multi-agent development, join us.

We’ll share the most important things we’ve learned while building a TypeScript runtime for concurrent AI agents, including:

- Reasoning vs. non-reasoning models
- How model context is structured
- Advanced agent-loop engineering techniques
- Event-driven coordination between agents
- Building agents that work concurrently without blocking one another

Join the hackathon: https://t.co/knsg01xYev

[Miodrag Vilotijevic @Mijuraaa]

What if agents were no longer trapped inside individual products?

PostHog, Figma, Notion, or any developer could build an agent that joins any project at runtime, becomes aware of the other participants, adapts to the system’s goals, and starts collaborating.

This would unlock entirely new ways to build, share, rent, and monetize agents. That is why agent interoperability could be the next big thing.

https://t.co/66MqyY0CgN

[Miodrag Vilotijevic @Mijuraaa]

Join a community of builders

The best way to learn AI engineering is not another tutorial.

It’s building something with other people who are experimenting with the same problems.

Our AI hackathon brings together developers interested in:
- AI agents.
- Multi-agent systems.
- TypeScript.
- LLMs.
- Agent architecture.
- Autonomous software.

You’ll build, exchange ideas, see how other teams approach the same problems, and become part of the growing Mozaik builder community.

September 5–6. Online.
https://t.co/XFQXBjPJkE

[Miodrag Vilotijevic @Mijuraaa]

Want to learn how to build multi-agent AI systems in TypeScript?

Join our online AI hackathon on September 5–6 and build a real system with multiple AI agents working concurrently.

You’ll get hands-on experience with:
- agent loops
- tool calling
- event-driven coordination
- concurrent agents
- human-in-the-loop patterns
- building autonomous AI systems

You’ll build with Mozaik, our open-source TypeScript framework for concurrent AI agents.

🏆 Today’s prize highlight: $500 for the winning team.

Join the builders → https://t.co/XFQXBjPbv6

[Miodrag Vilotijevic @Mijuraaa]

We’ve confirmed our hackathon partnership with @dailydotdev and @yourhyperskill .

📅 Online Hackathon — September 5–6

🏆 Cash prizes

🥇 $500
🥈 $300
🥉 $200

🎁 Subscriptions

10× annual @dailydotdev Plus subscriptions
5× six-month @yourhyperskill Premium subscriptions
5× six-month @mozaik_runtime Cloud Premium subscriptions

Register here: https://t.co/knsg01xYev

[Miodrag Vilotijevic @Mijuraaa]

Today’s multi-agent systems inherited the waterfall model: hardcoded handoffs, blocking dependencies, and constant waiting.

In this article, I explore the Agentic Waterfall Anti-Pattern, why parallel execution alone does not solve it, and how concurrent AI agents can open the door to faster, more resilient, and truly autonomous systems.

https://t.co/bdh5Cqs0lH

[Mozaik @mozaik_runtime]

There is no central scheduler or hardcoded workflow. Participants communicate through events and react as work unfolds.

Use it as a starting point, then add more agents and tools to the same environment.

[Mozaik @mozaik_runtime]

Mozaik makes it easy to build multi-agent systems where participants communicate, coordinate, and work in parallel inside a shared environment.

Here is a tiny example of how to build a CLI agent with Mozaik.

https://t.co/2sr2wJpjSv