NoKV x LingTai: design-partner collaboration for agent-native research workflows

[wchwawa]

Today we are starting a design-partner collaboration with LingTai on agent-native infrastructure for long-running scientific and research workflows.

NoKV is building a metadata control plane for object-backed agent artifacts: workspaces, logs, checkpoints, reports, and citable evidence in one filesystem-shaped namespace. LingTai is building local-first, filesystem-native infrastructure for resident agents: workdirs, mailboxes, memory, lifecycle, and inspectable state.

The overlap is concrete: real agent systems need durable context, not just another chat loop. Scientific and research workflows make that pressure visible because they involve heterogeneous data sources, long tool chains, provenance, cacheable intermediate artifacts, and resumable workspaces.

With LingTai as an early design partner, we will pressure-test NoKV around:

• workspace checkpoint and recovery semantics for long-running agents;
• artifact provenance and citable outputs;
• inspectable filesystem state for agent workdirs, mailboxes, logs, and knowledge;
• metadata interfaces that let agents query and resume work without reading an entire workspace;
• storage primitives for source-grounded, reproducible agent workflows.

This is a design-partner announcement, not an integration release. We are not claiming production deployment, institutional endorsement, or a finished API. The point is to bring real workflow feedback into the storage substrate early, while the system is still small enough to change.

If you are building artifact-heavy agent systems, scientific agents, or local-first agent runtimes, we would like to hear what metadata, checkpoint, and provenance primitives you need from the storage layer.

Useful links:

• NoKV: https://github.com/NoKV-Lab/NoKV
• LingTai: https://github.com/Lingtai-AI/lingtai
• LingTai Kernel: https://github.com/Lingtai-AI/lingtai-kernel