Pre-release · Built by working scientists
Science software, owned - not rented.
Gnometrix blends gnosis, deep understanding, with metrics, reproducible evidence. The Gnosis Suite is a set of local-first, AI-native tools for writing, statistics, microscopy, sequencing, lab records and research intelligence - designed to stay connected from experiment to publication, and to keep your data and your IP on your own machine.
In active development - nothing here is for sale yet. Email info@gnometrix.com with “beta-testing” in the subject.
The Gnometrix Suite
Science has a workflow problem. Gnometrix is built around one shared project context.
The tools that capture data, analyze images, run statistics, search literature, write grants and keep notes have never meaningfully talked to each other - so the researcher becomes the integration layer. Gnometrix closes that gap with purpose-built tools that share a common data model and one principle: every step from acquisition to publication should be reproducible, auditable, and fast enough to stay out of the way of the science.
The model strategy
Run it your way - and never on yesterday’s best model.
Every tool in the suite is built on a layered approach to language models. Start fully local for maximum privacy; step up to a frontier model only for the tasks that need it. The heavy scientific computation is deterministic - the model is an assistant, not the engine.
Maximum privacy, on your hardware
- Runs fully offline via Ollama - data and IP never leave your machine
- No API key and no per-token cost
- The right default for unpublished work and sensitive or regulated records
- For most tools a small model on an ordinary laptop is enough; the scientific computation is deterministic
Peak responsiveness when a task needs it
- One key, 300+ models, behind a single OpenAI-compatible endpoint (openrouter.com)
- Always the current best model - new frontier releases appear instantly, no migration, no waiting on one lab’s cadence
- Pay per token; idle costs nothing
- Reach for it for the harder reasoning: complex planning, multi-domain analysis, the toughest claim checks
Privacy & ownership
The paid API is more private than a chatbot subscription.
A ChatGPT Plus, Claude Pro or Gemini Advanced seat is not the private option. Routing paid APIs through OpenRouter - or staying fully local - protects your data and frees you from a single vendor.
| Dimension | Paid API / local (Gnosis approach) | Consumer chatbot subscription |
|---|---|---|
| Trains on your content by default | No - API input/output is not used for training | Often yes - opt-out required on most tiers |
| Data retention | Abuse-only, short retention; zero-data-retention option | Tied to your account; varies |
| Model choice | Any model, switch per call | One vendor’s lineup only |
| Cost model | Pay per token - idle = $0 | Flat monthly seat |
| Most private mode | Local Ollama - never leaves your device | Cloud-only |
Sources: OpenAI enterprise-privacy & data-use policy; OpenRouter privacy & zero-data-retention documentation. Policies change - verify current terms for your account.
The accuracy case
When a task is hard, combining models beats any single model.
Several tools can optionally cross-check their work across competing models. The published evidence for why that helps:
The tools
Purpose-built tools for the scientific workflow.
Each tool is in active development. Screenshots are from the working applications. Nothing here is for sale yet - email info@gnometrix.com to ask about beta-testing and availability.
GnoWrite
Replaces Word + EndNote / OverleafScientific manuscripts written from your own evidence — every claim checked, Word-native, private by default.
GnoWrite turns your own reference library into a submittable draft. A RAG index over your PDFs grounds the writing, a local claim-checker verifies each statement against the evidence you actually hold, and everything stays in real .docx with tracked changes, Word comments, Zotero/CSL citations and native equations.
- Evidence-grounded RAG over your own PDF library (GROBID + LanceDB, BGE-M3 / SPECTER2 retrieval)
- Local claim-checking: DeBERTa / SciFact NLI scores each claim against your evidence
- Real .docx round-trips — tracked changes, comments, cross-references, OMML equations
- Journal-aware: pulls current author guidelines before any reformatting
- Track-changed AI rewrites you approve or reject, with a deterministic caution fallback
GnoWrite runs its claim-checking locally: a cached DeBERTa / SciFact NLI model verifies claims against your own PDFs and an Ollama model handles rewrites. This is CPU-workable on a normal machine; a modest GPU mainly speeds the NLI and embedding passes. Opt-in frontier escalation through OpenRouter is available for the hardest claim checks - your choice, per task.
GnoGraph
An alternative to GraphPad PrismDescribe the experiment. Get the analysis and the publication figure — local-first, no subscription.
GnoGraph inverts the column-statistics-and-figure loop. Instead of hunting for a test in nested menus, you describe the experiment or paste a table; the local model routes it to the appropriate test, the statistics run deterministically, and the figure, assumptions, warnings and reproducible methods text are produced together.
- Experiment-aware routing from your table structure and design description
- Classical, non-parametric, mixed-effects and Bayesian models (SciPy, statsmodels, pingouin, lmfit, PyMC, arviz)
- Dose-response, enzyme kinetics, survival, curve fitting and model comparison (AIC/AICc/BIC)
- Volcano plots, pathway enrichment, PCA and spatial autocorrelation for omics and spatial data
- Reads GraphPad .prism / .pzfx as a re-analysis starting point, with provenance retained
GnoGraph is deliberately light on the model. The LLM only routes pasted data to the right test and writes the methods paragraph - the statistics themselves are computed deterministically by SciPy / statsmodels / pingouin / lmfit / PyMC. A very basic local model on an ordinary laptop, with no dedicated GPU, is enough.
GnoCyte
An alternative to Imaris / arivisNative visualization, segmentation and biophysics for large microscopy datasets — GPU-fast, open export.
GnoCyte brings large bioimage datasets, segmentation, tracking and quantitative biophysics into one local workspace. Two proprietary GPU engines — CytoD3D12 (Direct3D 12, the preferred engine on Windows) and Cytovulkan (Vulkan, cross-platform) — stream volumes too large for RAM, and results export openly rather than into a proprietary format.
- Lazy-loading of Imaris .ims (with scene objects), Zeiss .czi, Leica .lif, Nikon .nd2, OME-Zarr and OME-TIFF
- GPU 3D volume rendering — MIP, blended and Z-coded projections — via the CytoD3D12 and Cytovulkan engines
- Biophysics: FRAP kinetics, single-particle tracking, Number & Brightness, FCS, FLIM phasor analysis
- Object tables, kymographs, comparison plots, high-content plate workflows
- Open export (CSV, HTML, GIF, movie) and a localhost MCP assistant - no lock-in
GnoCyte is the exception in the suite: its hardware demand is the GPU rendering engine, not the language model. The renderers stream very large volumes at high frame rates and the biophysics runs natively in C++ - that is where the GPU goes. The AI assistant dock itself is light and runs against a small local model, or OpenRouter for richer protocol and report help.
GnoPyCyte
Companion to GnoCyteThe scriptable Python analysis engine underneath GnoCyte workflows.
GnoPyCyte exposes the same quantitative image-analysis routines as a reproducible, scriptable Python layer — for batch processing, custom pipelines and integration with the rest of your analysis stack.
- Scriptable access to segmentation, measurement and tracking routines
- Reproducible, batchable workflows with a workflows dock
- Plots and exports that match the GnoCyte desktop surfaces
- Designed to slot into existing Python analysis environments
As a Python engine, GnoPyCyte carries no model requirement of its own; any AI assistance is inherited from the GnoCyte desktop, where a small local model is sufficient.
GnoBio
Covers SnapGene + Geneious domains, plus NGSSequence engineering, NGS read workflows, spatial omics and a genome browser in one application.
GnoBio consolidates molecular biology and bioinformatics into a single local application. A local agent turns plain language into deterministic bioinformatics calls, and portable project bundles plus Git sharing mean no server ever hosts your sequence IP.
- Sequence management and construct engineering: primers, ORFs, Gibson assembly, codon optimisation
- NGS read workflows: FASTQ QC, trimming, mapping, variant calling via managed aligner binaries
- Spatial genomics: marker-based cell typing, neighbourhood enrichment, Ripley’s K
- JBrowse2-compatible genome browser and on-demand public reference data
- Portable .seqosbundle and Git-backed project sharing for reproducibility
GnoBio’s local agent is a lightweight tool dispatcher - it turns natural language into calls against deterministic bioinformatics routines. The default local model (Ollama qwen3:4b) runs on modest hardware with no GPU required for everyday dispatch. Route to OpenRouter when you want quicker, more capable planning across the 20+ analysis domains.
GnoBook
An alternative to Benchling / LabArchivesOffline-first, tamper-evident lab notebook for academic and regulated environments.
GnoBook is an electronic lab notebook built around immutability, offline operation and data sovereignty. Records are append-only and chained with cryptographic hashes, the client works fully offline, and everything can be self-hosted on your own infrastructure.
- Immutable records: changes produce versioned addenda enforced at the database level
- SHA-256 hash-chain audit log - a forensically verifiable, tamper-evident history
- Offline-first Flutter client (iOS, Android, tablet, web) with outbox/replay sync
- Compliance- and forensic-export ready for GxP / GLP / GMP requirements
- Self-hosted, air-gap capable - full data sovereignty
GnoBook’s AI co-pilot is an on-device Socratic reviewer (a Gemma-class model via a custom Ollama interface) plus deterministic CSV statistics. It runs on modest hardware and, by design, no data leaves your infrastructure - the right default for an append-only record and essential under GxP. External routing is reserved for explicitly non-sensitive help.
GnoScientist
Beyond generic AI research toolsA multi-agent research and grant studio — built by a working PI around the workflow generalist AI tools never reached.
GnoScientist encodes a PI’s actual workflow: autonomous deep research, gap detection, a grounded writing studio with an adversarial red-team, and a grant and resubmission engine. More than 80 specialist agents collaborate over a shared tool layer, and it augments the validators you already trust - scite, Consensus and Elicit - rather than replacing them.
- Autonomous deep research across PubMed, Semantic Scholar, bioRxiv and your own RAG-indexed library
- Gap detection that spawns new retrieval rounds until the evidence converges
- Writing studio with adversarial critics, a claim auditor and a statistics auditor on every claim
- Grant & resubmission engine: aim/significance framing, reviewer triage, point-by-point responses
- 80+ specialist agents (structural biology, genomics, pharmacology, clinical) over a shared MCP layer
GnoScientist is the most demanding tool in the suite, and we won’t pretend otherwise. For effective local use you need reasonable hardware - a capable GPU with the VRAM to run local models at a usable size and speed; a small CPU-only model can demo the workflow but won’t deliver the throughput 80+ agents need. For a highly responsive system at optimal performance, OpenRouter is the preferred path: it removes the local-hardware ceiling and powers the optional multi-model cross-validation.
GnoCyte rendering engines
Two GPU engines built for volumes too large for memory.
GnoCyte ships its own proprietary, validated 3D rendering engines — the hardware demand of the suite lives here, not in the language model. Both stream very large volumes at high frame rates and produce MIP, blended and Z-coded projections. The examples below are FIB-SEM reconstructions of a HeLa cell nucleus rendered with CytoD3D12.
CytoD3D12
The native Direct3D 12 engine and the recommended path on Windows — lowest-overhead access to the GPU for the smoothest interaction on very large datasets.
Cytovulkan
The Vulkan engine for portable, high-performance rendering across platforms and GPU vendors, with the same projection modes and validated output.
Licensing the rendering engines
Commercial entities interested in licensing the CytoD3D12 or Cytovulkan 3D rendering engines can contact us to discuss terms.
Free education
Open science education, built to the same standard.
Part of the Gnometrix portfolio is free education - rigorous, evidence-based resources for researchers, students and the public.
The Hypothesis Critic
Rigorous, evidence-based evaluation of landmark biomedical hypotheses.
A free education platform and YouTube channel that applies the Hypothesis Scoring and Evidence Framework (HSEF) - a structured rubric scoring hypotheses from α (strong causal evidence) to Ω (contradicted by high-quality evidence) - to landmark claims in cell biology, neuroscience, cancer biology, aging and immunology. The HSEF v3 prompt system runs through any LLM, local or cloud.
cellnucleus.com
An interactive cell-nucleus education resource — a separate Gnometrix education property.
cellnucleus.com is a standalone education site focused on chromatin and nuclear organisation, with interactive 3D visualisations and explainer reviews. It is hosted separately from this site.
Get involved
Interested in beta-testing?
Gnometrix tools are in active development and not yet shipping. We’re looking for working scientists to help shape them. To ask about beta-testing or product availability, email us and tell us which tools interest you, your operating system, and your role or lab type.
One address for everything: info@gnometrix.com