Honest availability on one DGX Spark — not a 250-model marketplace. Live runs today. Next / Roadmap follow a fixed ship order (cash path → ESMFold → DiffDock → one complex engine). Free taste is structure only — design and docking stay paid/managed.
FASTA → MSA → structure → pLDDT/PAE → QC-A path. Primary engine for free + Paid Pilot.
colabfold-af2One active heavy fold at a time. We pause competing GPU work during pilot folds.
/api/job/{id}Fast single-sequence folds — second PREDICT mode for orphans / snappy taste. Not Live until Spark smoke + QC.
CA-PROTEIN-PREDICT (engine=esmfold)Ligand pose hypotheses on a receptor we already folded. Paid add-on after structure — not free bait.
CA-PROTEIN-DIFFDOCKOpen AF3-class complexes after cash path is real. Board: Live at most one complex engine.
Parallel eval for multi-component + affinity-style scores. Lives only if it wins go/no-go vs OpenFold-3 — not both.
Inverse folding / sequence design when labs ask for design — paid DESIGN, never free attract.
Parked so we do not run three complex stacks. Install only if a pilot requires Chai by name.
Ship order and owners: staff Roadmap board (login).
Spec: docs/specs/candleai-engine-roadmap-execution-p0.md.