main

ADR-0001: Consolidate Local Inference Stories

Date: 2026-01-29 Status: Accepted

Context

We have five related user stories (001-005) covering local inference capability:

Story Title Dependencies
001 Local inference spike None
002 Hardware detection 001
003 Model download 001, 002
004 Local inference provider 001, 003
005 Default provider selection 004

Stories 002-005 have strict dependencies and only deliver value together. The spike (001) produces a different deliverable (research document + decision) than the implementation stories.

Decision

Keep story 001 (spike) separate. Consolidate stories 002-005 into a single implementation story.

Consequences

Positive:

  • Clearer deliverable: one story = one working feature
  • Spike/implementation separation follows established pattern
  • Reduces backlog overhead from 5 stories to 2
  • Easier to understand the end-to-end goal

Negative:

  • Larger implementation story may be harder to estimate
  • Less granular progress tracking during development