Open Source / Agentic Systems / Developer Infrastructure
A Rust-native, local-first agentic workspace harness and orchestration platform for secure AI-assisted workflows.
Conceptual system studyField notes
Giving a model access to tools does not resolve permissions, failed tasks, persistent context, worker isolation, human approvals, or the record of what happened. Those concerns become infrastructure as soon as an agent is allowed to act.
Karnelian brings identity, capability-based permissions, workers, memory, events, and auditing into a local-first orchestration layer. The aim is a programmable foundation where developers remain in control of execution.
System model
Produces reasoning and proposed actions.
Carries identity, task context, and intent.
Coordinates lifecycle, memory, tools, and events.
Constrains which capabilities may be exercised.
Perform bounded work across supported runtimes.
Local resources and approved external systems.
Conceptual flow — not a complete literal map of the current implementation.
Ecosystem
Terminal-based developer orchestration and multi-repository workspace management.
Active DevelopmentAgent execution, runtime coordination, security, memory, and tool orchestration.
Research RoadmapA long-term research direction toward an integrated agentic operating environment.
Engineering map
A systems-oriented foundation for coordinating agent execution.
Explicit, scoped permissions for every action an agent takes.
A defined identity and execution context for each agent run.
Context designed to remain useful across tasks and sessions.
Extensible worker execution across different runtime boundaries.
Event streaming and task coordination between system parts.
A traceable record of what agents attempted and performed.
A path for integrating models hosted within user-controlled infrastructure.
Development position
Karnelian is the lab's most developed public initiative. The CLI and Harness are active engineering areas; Karnelian OS remains a longer-term research roadmap rather than a released operating system.
Agent execution, security boundaries, workers, memory, and observable task lifecycles.
Multi-repository workspace control and local environment coordination through KRNL CLI.
Research into deeper coordination between agents, local models, and system services.
Explore what an agentic operating layer could eventually become.
Connected work
Open engineering
Follow verified development, review public code where available, or start a relevant technical conversation.