GENtle

Install · Five-minute tutorials · GUI manual · CLI manual · Agent interfaces · Showcase

GENtle is an open-source desktop and automation workbench for planning, simulating, inspecting, and documenting DNA cloning and sequence-context workflows. One deterministic engine powers the GUI, CLI, Python, JavaScript, Lua, MCP, and curated agent handoffs, so interactive decisions and automated runs share the same project state, reports, and provenance.

Both an inner agent in the running application and outer agents in separate automation environments are prepared to work with GENtle's routine catalog. They can discover routines, help bind their inputs, run deterministic preflight, request confirmation where required, execute approved steps, and return the resulting reports without creating a second implementation of the biology.

Internal preview: GENtle is already useful for research planning and reproducible in silico work, but interfaces and project files may still evolve. Inspect generated designs before taking them to the bench. The maturity map distinguishes recommended, caveated, and exploratory workflows.

Outer agent: reproducible artifact	Inner agent: live project context
An external agent can discover and run the headless workflow, then receive this SVG plus structured reports and provenance.	The in-application Agent Assistant works against the same active project while the user inspects its GRCh38.p14 anchor, evidence groups, and sequence map.

Both panels show local bases 1..1200 of the same 83,686 bp TP73 project. The current GUI renderer keeps transcript paths running to the viewport edge when their connected exon lies outside the visible span, so zooming no longer makes an intron look like an unexplained gap. Click either panel for the full-size view; exact generation and capture provenance is recorded in the figure catalog.

A minimal source-build check is:

cargo check -q
cargo run --bin gentle_cli -- capabilities
cargo run --bin gentle

For a headless CLI/MCP build that excludes the desktop GUI dependency stack:

cargo build --locked --release --no-default-features \
  --bin gentle_cli --bin gentle_mcp

The default feature set continues to build the desktop application.

An existing RNA mapping report can feed its accepted target-gene reads directly into the allele-aware evidence screen:

cargo run --bin gentle_cli -- --project project.gentle.json allele-hash-screen \
  --gene FUS --from-rna-report REPORT_ID --transcript-fasta transcripts.fa \
  --variant-table variants.tsv --min-informative-reads 10 \
  --balanced-band-lo 0.40 --balanced-band-hi 0.60 --out allele-screen

Salmon unmapped_names and mapping SAM inputs can select additional cohorts, but still require the original FASTA/FASTQ via --read-file or --read-pair. The v3 report adds depth-aware hap1_preferred, hap2_preferred, balanced, inconclusive_low_depth, or unphased_allele_level_only sequence- representation verdicts. These are threshold-based evidence summaries, not claims of biological significance or causation. See the CLI manual for the complete contract.

Tagged internal releases may provide a macOS DMG or Windows ZIP on the releases page. Linux users can build from source or use the documented container route. See INSTALL.md for platform details and optional external tools.

• One engine, many interfaces. GUI actions, shell commands, workflows, scripts, and agent tools converge on shared biological operations.
• Project state rather than disconnected files. Sequences, annotations, containers, arrangements, reports, lineage, and imported evidence remain linked.
• Explanations from the computation. Protocol cartoons, lineage graphs, dotplots, expert views, and structured reports are derived from the same state that produced the result.
• Genome context when cloning needs it. Prepared references, genome anchors, GFF/GTF annotation, BED/BigWig/VCF tracks, repeats, motifs, and RNA evidence can inform the same project.
• Reproducibility by construction. Operations and results are machine-readable, deterministic where inputs are fixed, and exportable for review.
• Biological adjustments by composition. Related assay and interpretation requests can reuse engine-owned transcript, coordinate, candidate, scoring, evidence, history, and reporting services instead of starting again in a frontend or one-off script. The extension guide shows contributors and coding agents how to add only the missing decision.

Designing a pair that detects transcripts is not the same as distinguishing their isoforms. GENtle combines explicit transcript scope, assay objectives, source-bound evidence and primer/product constraints, then keeps coverage, discrimination and readiness checks inspectable.

This maintained explanatory diagram uses conceptual, not computed, transcripts and assay outcomes. Optional inputs are not automatically available or required by every operation; candidate, assessed and policy-qualified order-ready do not mean experimentally validated. Open the full-size vector graphic, read its source map and interpretation notes, or follow the PATZ1 primer-pair tutorial.

These labels describe implementation maturity, not biological validation. Users remain responsible for reviewing designs, assumptions, and experimental conditions.

The detailed release gate and open work live in the roadmap. Completed changes are recorded in the changelog.

Interactive interfaces and imported biological context meet in one shared engine. The resulting project state drives cloning, retrieval, design, analysis, rendering, and provenance rather than leaving each frontend to reimplement the biology.

Use an operation when the exact atomic step is known, a routine when GENtle should help bind and explain a workflow, and a specialist when the task benefits from focused interactive review.

The full scientific and technical gallery is in the GENtle showcase. Exact figure-generation commands and asset provenance are maintained in docs/figures/README.md.

GENtle can plan and apply a destination-first Gibson assembly, retain explicit overlaps and primer suggestions, reopen the operation from lineage, and render the modeled mechanism. The same state can produce protocol cartoons, lineage graphs, container arrangements, rack layouts, and gel readouts.

See the complete cloning and physical-workflow examples.

Prepared references and genome anchors let GENtle compare transcript isoforms, splice structure, repeats, interval tracks, microarray evidence, TFBS support, and nucleotide sequence without losing assembly or coordinate provenance.

See TP73 and TERT genome-context examples.

The shared PCR family covers endpoint PCR, advanced PCR, mutagenesis, selection-first primer-pair design, and probe-bearing qPCR assay design. Reports preserve sequence constraints, candidate diagnostics, products, and links to later oligo-order planning.

See the PCR, qPCR, and TP73 isoform-selector examples. For a feature-by-feature boundary with NCBI's hosted service, see GENtle and NCBI Primer-BLAST.

MCP exposes typed tools plus capability metadata to compatible agent hosts. ClawBio/OpenClaw uses a narrower curated request/result wrapper. The VKORC1 example turns a pharmacogenomic prompt into an inspectable promoter-fragment and reporter-planning bundle without bypassing confirmation or provenance.

See the complete agent-handoff story.

First-class buttons and named tools differ by interface, but all interfaces retain a path to the shared engine. The GUI/CLI/MCP parity matrix distinguishes engine reachability from intentionally curated presentation.

Inner agent: the in-application Agent Assistant. It receives the active project summary and works through shared agents, routines, shell, operation, and workflow paths. The user chats inside GENtle; the agent may ask for missing information, compare routines, or offer reviewed commands for confirmation or permitted execution.

Outer agents: separate automation environments. Codex, Claude-style coding agents, MCP hosts, ClawBio/OpenClaw, and similar processes use MCP capability discovery, CLI help/capabilities, the shared shell, or curated skill requests. The external environment discovers what GENtle can do, supplies project/state context, invokes the same routine and preflight machinery, and collects structured artifacts.

The routines themselves are prepared and described by GENtle, including the typed catalog in assets/cloning_routines.json and shared routines list, routines explain, routines compare, template-binding, and execution routes. Neither kind of agent is expected to automate the GUI by guessing where to click. Both operate through these shared descriptions and commands, and both retain confirmation boundaries for mutating operations, external handoffs, and purchases. A reply that only asks a question or chats does not imply that anything is ready to execute.

The Agent Assistant and agent interfaces tutorial shows who runs what, where capability discovery happens, and how inner and outer agents differ from direct CLI use.

The core GUI, CLI, MCP server, and documentation paths need only Git and a Rust toolchain for a source build. JavaScript and Lua are optional Cargo features; the Python package is a separate CLI-backed integration.

Advanced workflows can use BLAST+, Primer3, ViennaRNA/RNAPKIN, bigWigToBedGraph, and external legacy SHA-1 verification tools when present. None is required for the first GUI launch. Installation, executable overrides, containers, and platform notes are centralized in INSTALL.md.

• Current package version: 0.1.0-internal.11. This alpha candidate awaits exact-revision package checks, Glen's readiness verdict and owner approval.
• Active release story: gene-informed primer-pair studies, source-aware isoform comparison, TSS collection workflows and auditable annotated/vector exports.
• Default builds include GUI, CLI, MCP, and documentation paths.
• Native release packaging enables JavaScript and Lua; the container is headless, without GUI or embedded scripting runtimes.
• Generated showcase figures come from GENtle engine outputs and/or versioned deterministic repository tooling. Maintained explanatory diagrams are explicitly labelled and are not analysis outputs.

See the roadmap for the current acceptance gate, the release guide, the v0.1.0-internal.11 release notes, and the complete release-note index for notable user-facing changes.

• One deterministic core: biology and workflow logic belongs in the engine, not in GUI, CLI, Python, JavaScript, Lua, MCP, or agent adapters.
• Provenance by default: derived results should remain traceable and replayable.
• Machine-readable behavior: operations, results, errors, readiness, and capabilities should be structured as well as human-readable.
• Thin, purposeful interfaces: each interface may emphasize different tasks without inventing different biological behavior.
• Human confirmation at consequential boundaries: agent and service handoffs must not silently turn suggestions into experiments or purchases.

GENtle is distributed under GPL-2.0-or-later. Bundled third-party data may carry additional attribution or redistribution terms; see copyright for the authoritative file-by-file record.

Project history and contributors are recognized in ACKNOWLEDGEMENTS.md.