AI Scaffolding
Local-first prototype featuring task normalization, model routing, worker roles, a critic and validated delivery contracts.
Problem
Initial situation
Complex tasks need multiple processing steps, different model roles and verifiable handoffs instead of a single opaque model response.
Constraints
Boundary conditions
- Integrate local models
- Explicit roles
- Observable states
- Interchangeable components
Solution
Solution
AI Scaffolding orchestrates model roles via Ollama. The Task Runner processes tasks with state tracking and resumption; specialized workers handle text, vision, retrieval, editing and code. Automatic task decomposition and integration with Nea Polis are identified as future extensions.
Architecture
Architecture
- Python / Ollama
- Task normalization / semantic routing
- Stateful sequential task pipeline
- Worker / Critic / Validators
- RunResult / manifests / traces
Result
Result
The specified pathways are present in the inspected repository revision. No execution with local models and no new quality or security certification; the README identifies the status as Early Release.