Alchemist API & Automation

Alchemist: Pipeline Permutations

Reference for the pipeline shapes a ticket can run through.

| View as Markdown
Hunter Hodnett
Hunter Hodnett CPTO at Chipp
| 1 min read
# alchemist # platform # pipeline-permutations

The Alchemist agent pipeline is content-shape, not time-shape. The two independent axes the planner is chosen on, and the model that fills each slot, are pinned in docs/pipeline-permutations.md (the design record). This page is a short in-product pointer to that doc so the model-preferences panel’s Learn more link lands somewhere real.

Tier × slot

  • Plan — DEFAULT planner is glm-5.2 (the high rung), routed via OpenRouter. max tier / explicit request upgrades to gpt-5.5 (the measured plan winner). fast drops to grok-4.3. economy (explicit only) drops to deepseek-v4-pro.
  • Implement — DEFAULT is minimax-m3 (top fix-quality in the faithful implement eval at ~1/25th gpt-5.5’s cost). max tier upgrades to gpt-5.5; trivial falls through to Sonnet.
  • Review / security / reconcile / merge_resolution — frontier (Opus 4.8).
  • Docs — Sonnet 4.6 (1M context covers the corpus + diff at lower cost).
  • Investigate — Haiku unless max tier (which upgrades to the frontier).

cheap tier forces Haiku across the board and skips the docs pass entirely. See Agent pipeline configuration for the per-project override surface, and the in-product Model preferences panel for the per-tier caps.