Memory Meets Motion: Our First LaserData Hackathon

Today in San Francisco, we are co-hosting Memory Meets Motion, our first LaserData hackathon.

View event details and register on Luma →

---

The challenge is simple to describe and difficult to get right: build an agentic workflow that moves with memory, state, and real-time data.

The current generation of AI applications has made prompts feel like the center of the system. But a prompt is only one moment in a much larger process. Useful agents need to receive events, call tools, share context, remember what happened, coordinate with other agents, recover from failures, and explain the decisions they made.

Agents need more than prompts. They need a nervous system.

That is what this hackathon is about—and it is the layer we are building at LaserData.

From prompts to durable agentic systems

The core question for every team is: how does data actually move between agents?

A production agentic system needs to do four things well:

• Move: carry events, messages, telemetry, and tool calls in real time.
• Remember: preserve long-lived context and durable memory beyond a single model invocation.
• Coordinate: manage multi-agent handoffs, retries, deadlines, and workflows.
• Replay: make it possible to debug, audit, learn from, and recover past execution.

These capabilities are closely related. A memory system without motion becomes stale. A stream without state cannot make durable decisions. A multi-agent workflow without replay is almost impossible to debug. The interesting systems emerge when all four work together.

We want Memory Meets Motion teams to build those systems—not another isolated chatbot, but an application whose intelligence persists and evolves as data moves through it.

The foundation: Apache Iggy

LaserData is built on Apache Iggy (Incubating), a Rust-native streaming engine designed for persistent, hyper-efficient data movement and ultra-low tail latency.

Iggy uses a thread-per-core, shared-nothing architecture, modern Linux io_uring, and a zero-copy hot path to make performance predictable. It is the durable foundation beneath the rest of the LaserData stack.

The project has now passed 500,000 downloads across Cargo, Docker, and its SDKs, attracted more than 100 contributors, and earned more than 4,500 GitHub stars. It is also one of the most popular and active Rust projects at Apache.

For hackathon builders, the architectural details translate into something practical: a fast, durable log that can carry the continuous flow of observations, actions, memory updates, state transitions, and agent-to-agent messages.

Eight primitives, one grammar

The Laser SDK turns that streaming foundation into developer primitives for real-time AI applications. Every primitive follows the same shape: one accessor, one verb, and an asynchronous result.

object.verb(input).await

The eight primitives cover the core data needs of an agentic system:

• Log for publishing, consuming, batching, and replaying records by offset.
• Memory for remembering, recalling, and consolidating durable knowledge.
• Context for assembling a conversation record and scoping memory to it.
• Graph for linking entities, traversing relationships, and finding nearest vectors.
• State for point reads and writes, compare-and-swap, leases, and copy-on-write branches.
• Changes for waiting on a view to advance instead of polling it.
• Views for filtering, aggregating, paging, and vector-searching projections.
• Fabric for directed agent requests, deadline contracts, ordered workflows, and runs.

Together, these primitives give teams one consistent grammar for streams, memory, state, context, and workflows. The Laser Stack then provides a durable path from an application to external stores and runtimes—including PostgreSQL, Elasticsearch, Iceberg, Doris, HTTP, RocketRide, FalkorDB, and others—without adding point-to-point wiring every time the system grows.

Teams can start locally, use LaserData Cloud for a prototype, or move toward managed, BYOC, and on-premises deployment models as the project matures.

What to build

The strongest Memory Meets Motion projects will make multi-agent collaboration, low latency, and ecosystem integration visible in the demo.

Some starting points from our kickoff:

• A voice agent with memory that streams calls, updates what it knows, and can replay the decisions it made.
• A streaming RAG system with fresh context, continuous indexing, and low-latency retrieval.
• A multi-agent workflow in which a planner, researcher, executor, and reviewer coordinate through durable state.
• An agent observability system that audits tool calls, state transitions, retries, and failures.
• A real-time support copilot that combines live events with customer context to recommend the next action.

These are prompts, not boundaries. The rule of thumb is simpler: if the demo has durable movement of data, it is probably a great fit.

Our team will be available throughout the hackathon to help with API keys, SDK integration, architecture reviews, connectors, and demo support.

Special LaserData prizes

In addition to the main hackathon awards, we are offering $2,000 in special LaserData prizes:

• $1,000 — LaserData Real-Time AI: best overall project built on LaserData.
• $500 — LaserData Multi-Agent: best collaboration between autonomous AI agents.
• $500 — LaserData Durable Intelligence: best use of streams, memory, context, and workflows to build a resilient AI system.

Every participating team that builds with LaserData will also receive cloud credits, direct access to the founders, and support amplifying the project after the event.

Build something that moves

Memory Meets Motion is a chance to explore what comes after the prompt: systems where intelligence is not trapped inside one request, but can move through a durable stream, accumulate context, coordinate work, and recover when reality gets messy.

That is the future we want to help builders create. We cannot wait to see what ships.

Start with the Laser SDK on GitHub, read the Laser SDK documentation, and join the Real-Time & AI Data Infrastructure Discord for help during the hackathon.