AI agents today are powerful, but still fundamentally stateless and fragmented. They complete tasks in isolation, lose context across sessions, and struggle to share knowledge across tools, teams, or workflows. Memory is often tied to a single app, model, or device — making agent systems brittle, hard to scale, and difficult to trust.

As agents evolve from simple assistants to autonomous, long-running systems, they need a more durable foundation:

• the ability to store and retrieve memory across sessions
• share context across agents and workflows
• and access data that is portable, persistent, and not locked into a single platform

This track challenges you to rethink how agentic systems are built by using Walrus as a Verifiable Data Platform for AI.

What you’ll build

Build functional AI agents or agentic workflows (single or multi-agent) in any domain — from finance to productivity to gaming — that demonstrate:

• Long-term memory using persistent, verifiable memory for agents
• Persistent data and file access using Walrus (directly or via a file management interface)
• Integrations and tooling that make it easier for developers to adopt Walrus or MemWal (Walrus Memory) in agentic systems

To guide you, we’re especially interested in:

• Long-running workflows where agents track state over time (e.g., research agents, trading agents, monitoring systems)
• Multi-agent coordination, such as negotiation, task delegation, or step-by-step execution across agents
• Artifact-driven workflows, where agents generate, store, and reuse files like datasets, logs, reports, or intermediate outputs

For integrations and tooling, think along the lines of:

• adding persistent memory to existing agent frameworks or tools (e.g., plugins or adapters to use Walrus directly, or to use MemWal as the Walrus Memory layer)
• creating workflow orchestration layers that combine memory, messaging, and execution across agents with Walrus as the underlying storage foundation
• enabling cross-tool or cross-agent memory sharing, where different systems can read/write to the same context stored on Walrus
• building interfaces or developer tools that make it easier to inspect, debug, or manage agent memory and data stored on Walrus