Agent Systems Stack
Open-source work on agent context, runtime, and collaboration.
These repos are meant to fit together as one stack, not as isolated experiments.
The Stack
| Layer | Project | What it is |
|---|---|---|
| Data plane | Nexus | Filesystem/context plane for agents. Shared paths, memory, storage, and connectors behind one interface. |
| Runtime | Koi | Agent operating system. Declarative agents, tools, channels, middleware, governance, and safe self-extension. |
| Collaboration | Grove | Contribution graph for asynchronous multi-agent work: contribute, review, reproduce, and adopt without forcing everything through one branch. |
How It Fits Together
Nexus gives agents a shared world to read, write, and search.
Koi runs agents inside that world with stronger safety, orchestration, and control.
Grove coordinates work across many agents, artifacts, and parallel lines of investigation.
Nexus -> shared files, memory, and context
Koi -> execution, governance, and orchestration
Grove -> collaboration, review, and adoption
Where To Start
- Start with Koi for the runtime model and overall architecture.
- Try Nexus for the most concrete hands-on entry point.
- Read Grove for the collaboration model and contribution graph.
Current Status
All three projects are early and evolving quickly.
- Nexus is in beta and already usable as a local or daemon-backed filesystem/context plane.
- Koi is an active runtime prototype and architecture-heavy buildout.
- Grove is in Phase 1: protocol plus local implementation.
What I Care About
- Durable memory and a coherent data plane for agents
- Safe, inspectable runtimes instead of black-box autonomy
- Collaboration through explicit artifacts and graphs, not hidden state
Feedback
Issues, ideas, and experiments are welcome.
Open an issue in the relevant repo if something is unclear or broken.