High-level summary of the AI Fabrick — goals, capabilities, and outcomes.

Overview

Codmir AI Fabrick turns Codmir from a project/ticket interface into a generalized agent fabric. The fabric coordinates role-specialized agents to plan, execute, QA, and deliver work.

  • Differential agents: roles imitate real-world responsibilities (PM, Eng, QA, Ops, Sec).
  • One prompt → crew plan: a planner maps the prompt to a small DAG of roles.
  • Tier routing: Router decides light vs heavy execution per step.
  • QA gates: Proctor validates outputs (schemas, safety, quick tests) before Ready.
  • Continuations: Long tasks yield/resume across function time limits.
  • Audit: All decisions and events are recorded for transparency.
flowchart TD
    U[User Prompt] --> O[Overseer]
    O -->|Route| R[Router]
    R -->|Local| L[Light Worker]
    R -->|Heavy| H[Heavy Worker (RunPod)]
    L --> P[Proctor]
    H --> P
    P -->|Pass| Ready[Result: Ready]
    P -->|Fail| Escalate[Rework / Escalate]
    Escalate --> R