Every tool, grouped by what it does.

Constrained combinatorial generation over real, controllable parameters: regulatory context, codon target, guide choice. Every candidate is scored by the predictors and run through the validators and biosecurity gate before it is ever proposed. A blocked candidate is dropped, not ranked.

One free-text instruction, decomposed into an ordered tool plan and chained end to end, e.g. express this, then predict its yield. The deterministic intent router classifies modality, method and delivery first, so the plan is auditable.

A plain-English explanation of any design where every claim is backed by an engine output or a citation. The LLM is hard-constrained to restate only the facts it was given, so it cannot introduce an ungrounded number.

Knockout / knock-in across Cas9 (NGG), Cas12a (TTTV), λ-Red, integrase, base & prime editing. Host-universal delivery, a CFD-style genome off-target screen, an HDR donor with re-cut prevention, and verification primers.

HGVS-lite edits like S65T, F64L, K27del: frame-preserving, with primer design and a re-check of the resulting construct. The exact deterministic move a bench scientist would make, automated.

Gibson, Golden Gate, MoClo and BioBrick RFC[10], plus a modern suite: In-Fusion, Gateway BP/LR, TOPO, restriction-ligation. Both-strand domestication scan so no method-breaking site slips through.

N/C-terminal tags, linkers, NLS/NES localization signals, secretion leaders and protease sites, composed and validated as one construct, with the junctions checked.

NNK/NNS site-saturation and combinatorial promoter × RBS libraries, with the diversity math worked out: theoretical size, codon coverage and the screening burden you're signing up for.

Operons, pathways and toggle-switch circuits: multiple coding sequences with their own regulatory parts, assembled and balanced into a single plannable construct.

Live, continuous QA: frameshift, premature stop, cryptic ORF, illegal restriction site, low-complexity/repeats, splice site, polyA signal, all returned as structured findings with severity, so a UI can underline problems as you type, exactly like a code linter.

Needleman–Wunsch (global) and Smith–Waterman (local) alignment, plus Sanger trace reading: confirm an edit landed and read the chromatogram without leaving the workspace.

Tm, hairpin, self- and cross-dimer, and specificity for any primer set, flagged before you order, not after a failed reaction.

A fast offline foldability heuristic runs in-loop on every Express design; a deep AlphaFold / ESMFold prediction with per-residue pLDDT confidence bands is available on demand.

ORF and feature analysis plus render-ready map tracks for a SnapGene-style plasmid view (GC ring, bp ruler, directional feature arrows, single-cutter enzyme ticks), drawn client-side with no libraries.

A precise design diff between two constructs or two versions (what changed, where, and why), feeding the version-control timeline and the side-by-side comparison report.

A CFD-style specificity prediction for a guide against the host genome: value plus confidence interval, with the evidence sites it scored.

Cutting efficiency for a chosen guide, scored with the same uncertainty contract, so you rank guides on evidence, not vibes.

Reference-model soluble yield in mg/L with a transparent factor breakdown (codon adaptation, GC, host, regulatory strength), so you see why, not just a number.

Multi-objective ranking with a true Pareto front: surface the non-dominated trade-offs (yield vs. specificity vs. cost) instead of collapsing them into one fragile score.

An in-house bench-results store with CSV/JSON ingest, a predicted-vs-observed reconciliation, and a closed-form least-squares calibration tracker (with R²), so the predictors get sharper as your data grows.

UCB / EI / uncertainty acquisition with a diversity penalty. It picks the batch that teaches you the most per dollar, not just the highest-mean candidates.

Git-like repos for constructs: branch, commit, log, diff, blame, with per-base attribution. Treat a plasmid like a codebase, with a real history you can audit.

The container that holds runs, repos, campaigns, sessions, data and files together, with an activity log and inline named sequences. A tabbed workspace: Overview · Repos · Campaigns · Chats · Data · Files · History.

A deterministic NL edit-and-re-check: "swap the promoter for EF1α and re-optimize" applies the edit, then re-runs the linter, analysis, foldability and yield automatically.

One-way, opt-in export to your ELN. Quill's GenBank output opens directly in the sequence tools you already use: design-intelligence in, system-of-record out. It never ingests from or depends on an ELN.