The AI Context Problem
For years, we've treated AI coding assistants like transient chat partners. You open a window, paste some code, ask a question, and get an answer. When you close the window, the context builds up, leaks, or vanishes.
This is fundamentally broken for engineering.
Complex software engineering requires state, memory, and rules. A chat window has none of these in a rigorous sense. It relies on probabilistic context retrieval, which is why AI often hallucinates imports, forgets previously established patterns, or rewrites code it just wrote.
Enter Project-Local Intelligence
ORKA shifts the paradigm. Instead of the AI living in the cloud or the IDE sidebar, the intelligence lives inside the project folder itself.
-
1
Deterministic Skills
Skills are not prompts. They are mini-programs stored in
.agent/skillsthat dictate exactly how to build components. -
2
Persistent Brain
The project remembers its own architecture in
.orka/brain.md. It knows what it built yesterday. -
3
Orchestration over Generation
ORKA doesn't just spew code. It plans, verifies, and updates its internal state before moving to the next task.
AI: *Guesses auth provider*
AI: *Invents file structure*
AI: *Forgets to update .env*
ORKA: "Reading .orka/brain.md..."
ORKA: "Project uses NextAuth."
ORKA: "Loading auth-module skill..."
ORKA: "Plan created: Add Google Provider."
ORKA: "Executing..."