Tomo Pipeline — Reducer and Effect Driver

A pure state transition chooses each bounded effect until one terminal intent settles the run

01 / Run Admission 02 / Tomo Turn Engine 03 / Effect Driver 04 / Shared Protected Host 05 / Durable Result Admit Reduce + execute Observe + settle Accepted Input member objective Run Snapshot identity + budgets Reduce State pure transition Next Effect typed intent Execute Effect model, tool, compact, checkpoint Return Observation result or failure Reduce Again continue or terminal Settle Once journal + result terminal bounded request committed outcome Legend User UI Agent logic Policy Tool action Context / trace

Tomo Control

  • • Reducer alone chooses the next effect
  • • Driver feeds every observation back into state
  • • One terminal intent ends the run

Best-Fit Tasks

  • • Direct answers and grounded retrieval
  • • Predictable single-objective turns
  • • Short, bounded tool chains

Shared Protection

  • • Host rechecks authority and budgets
  • • Tools, checkpoints, journal, and settlement stay shared