Karan Jagtiani
Founder of Skyflo. I build agentic systems, AI infrastructure, and developer tooling.
press ~ for the terminal
What I'm building
The mission is the unit of work.
Skyflo is an agentic engineering harness. One objective becomes a durable mission: it reads the systems the work will touch, plans that work for your approval, then delegates to isolated specialists that execute across code, cloud, CI/CD, and operations.
A separate agent reviews the outcome, and the context worth keeping carries into the next mission.
Skyflo
AI engineering harness · Desktop for macOS
A mission holds
Where the limits are
The standalone Kubernetes agent stays open source under Apache 2.0. It is a separate project from the Skyflo harness.
117skyflo-ai/skyfloWhere I spend my time
Four layers of the same problem.
Agentic systems
Mission orchestration, delegation, and the control loop that keeps specialist agents pointed at one objective.
- surface
- missions · specialists · approvals
- problem
- delegation boundaries
AI infrastructure
The runtime agents execute inside: model routing, context handling, and keeping the work on the user's own machine.
- surface
- runtime · routing · context
- problem
- local execution
Engineering platforms
The durable state behind a mission that outlives the session: workspaces, history, automations, retained context.
- surface
- repos · terminal · browser
- problem
- state that persists
Harness design
The product surfaces engineers operate through, and the approval model deciding what an agent is allowed to do.
- surface
- Skyflo Desktop for macOS
- problem
- approve before act
The interesting problems sit between them. That gap is exactly where an agent has to earn the trust to keep going.
Open to a conversation
Say hello.
Working on agentic systems, AI infrastructure, or developer tooling? I'd like to hear about it.
Usually replies within a day or two.