Open Source / Karnelian
Karnelian Harness — Structure for autonomous execution.
The Karnelian Harness investigates the systems required to make AI-assisted execution more structured, inspectable, and controllable.
Agents need more than access to a language model. Useful execution requires tools, boundaries, runtime coordination, contextual memory, approval mechanisms, and observability.
Conceptual system studyTechnical panels
Inside the harness
Runtime security
Capability-based boundaries define what an agent may touch, and nothing more.
Workers
Execution across multiple runtimes, coordinated by the harness.
Memory
Contextual memory that persists across tasks.
Task lifecycle
Structured states from request to completion, including failure midway.
Approvals
Human approval mechanisms for sensitive actions.
Auditing
Observability into what an agent attempted and what it did.
Ecosystem
Karnelian initiatives
KRNL CLI
Terminal-based developer orchestration and multi-repository workspace management.
Active DevelopmentKarnelian Harness
Agent execution, runtime coordination, security, memory, and tool orchestration.
Research RoadmapKarnelian OS
A long-term research direction toward an integrated agentic operating environment.