NVIDIA Inception Premier member · built on NVIDIA RAPIDS & CUDA
What is LynxKite?
Agentic, multi-scale modeling
LLM agents orchestrate open and proprietary models — from genes to molecules.
GPU-accelerated at scale
RAPIDS/cuGraph under the hood for graphs that don't fit on a laptop. Your open-model GPU microservices start and stop with the workflow, so idle GPU time does not land on the programme budget.
Graphs to drug candidates
Graph reasoning, ML, and molecule analysis in one workflow.
Open core, enterprise-ready
Free and open source, with a commercial tier for production teams.
// The workspace
Watch your pipeline build, live.
This canvas is live — click any box to inspect it, then hit Run on GPU and watch the orchestrator execute the pipeline stage by stage, spinning GPU microservices up and down automatically.
compound-screening.lkgnn-train.lk
Knowledge graphSOURCE
4.2M nodes
RetrainLOOP
Reinforce & retrain
GNN ModelTRAIN
Property prediction
BioNeMoGENERATE
Compound design
RDKit filterCHEM
Tanimoto ≥ 0.7
VisualizeOUTPUT
Ranked candidates
● GPU node up · auto-scaled
Life Sciences
Built for computational biology and drug discovery
2D/3D molecule visualization, chemoinformatics, and integrations with NVIDIA BioNeMo and RDKit bring GPU-accelerated, AI-guided molecule generation and screening — not bolted on, built in.
In production
LynxKite powers Biophytis' AI-driven longevity drug-discovery platform — the MASSIVE project — showcased with Biophytis, Nebius, and Accenture at NVIDIA GTC 2026.
Run any Hugging Face model as a LynxKite Inference Microservice
Biology moves faster than packaging. A LIM turns a plain Python row function into a scaling GPU service on Kubernetes with a single @lim decorator — no images, no manifests, no waiting on a vendor container. The whole Hugging Face Hub becomes callable from a box users already know.
Sequences in as rows, predictions back as columns — the same table, plus one new column.
LynxKite owns deploy, scale and retries — your model logic stays in Python, unchanged from the Hub's example code.
Sits next to NVIDIA NIMs in one GPU-service list — and resizes to zero replicas when no workspace needs it.
An example drug-discovery workflow, exported straight from LynxKite and running right here in the page. It generates candidate molecules with REINVENT4, screens them with rule-of-five, synthesis, solubility and ADMET filters, docks them against the MAS1 receptor structure, and keeps the best binders. Click any node to explore.
Where targets come from
Graph neural networks, plus GraphRAG and LLM reasoning over literature, regulatory text, omics and lab data, surface candidate targets. Click a node to see what it represents.
Gene
Click any node in the diagram above to see what it represents.
Open core, by design
Open Source
LynxKite 2000:MM is open source. Install it, run it, extend it.
A commercial license adds cluster deployment, additional enterprise features, and dedicated support for production teams. A proprietary commercial license keeps partner code, models and deliverables free of copyleft obligations.
Graph reasoning and GPU-scale compute from discovery to preclinical.
Pharma & Life Sciences
Target Identification
Graph neural networks and GraphRAG with LLM reasoning over a biomedical knowledge graph rank candidate targets, with the evidence trail attached.
In Silico Compound Screening
GPU-enhanced generative compound design, optimization, and screening with NVIDIA BioNeMo and RDKit built in.
PK/PD Modeling
AI-powered pharmacokinetic and pharmacodynamic models describing absorption, distribution, metabolism, and elimination.
Heritage: 100+ enterprises, including Roche and AstraZeneca.
Built to fit your stack
Plug into the ecosystem you already use — no rip-and-replace required.
NVIDIA BioNeMo
Pre-trained biomolecular foundation models for molecule generation & analysis.
NVIDIA cuGraph
100+ GPU-accelerated graph algorithms via NVIDIA RAPIDS.
RDKit
Industry-standard cheminformatics toolkit integrated directly into workflows.
Custom LLMs & APIs
Bring OpenAI, Anthropic, or your own models into agentic workflows.
Vector & Graph DBs
GPU-accelerated, enterprise-ready database connectors out of the box.
Cloud-native deployment
Kubernetes, AWS, Azure, or GCP — containerized, GPU-ready, scalable.
Technology partners & programmes
NVIDIA Inception Premier
NVIDIA Cloud Validated
Nebius AI
Amazon AWS
Google Cloud
Watch LynxKite 2000:MM In Action
See the platform live — from pipeline design to GPU inference.
Build Custom Boxes
Extend the platform by writing and registering your own reusable operations directly in Python — no forking required.
Frequently asked questions
Everything you need to know before getting started.
General
LynxKite 2000:MM is an AI platform for building, deploying, and scaling graph-native analytics and machine learning workflows, visually or in Python. It helps teams turn complex, connected data into actionable insights.
No. You can build workflows using a drag-and-drop interface, making advanced AI and graph-based analytics accessible to non-programrs. Power users can also drop into a full code editor and extend everything in Python.
Yes. LynxKite 2000:MM is designed for interoperability. You can integrate your own models, connect to cloud storage or enterprise databases, and incorporate external data sources into your workflows.
LynxKite 2000:MM is built to model and analyze relationships in your data using graph structures. It supports Graph Neural Networks (GNNs), knowledge graphs, and graph-based reasoning — making it ideal for domains where connections between entities matter.
The platform runs on GPU-accelerated infrastructure and supports large-scale AI workloads, from graph training to Retrieval-Augmented Generation (RAG). With 100+ cuGraph-accelerated algorithms and automatic GPU microservice management, it handles millions of nodes efficiently.
The platform offers multi-user collaboration with shared workspaces. Engineers, analysts, and domain experts can co-develop workflows, models, and visualizations in real time — increasing productivity and reducing handoff delays.
For Data Engineers
LynxKite 2000:MM integrates with GPU-accelerated vector and graph databases, custom LLMs and APIs, open-source toolkits like RDKit, cloud-native services (AWS, Azure, GCP), and enterprise systems. It also supports NetworkX-compatible graph algorithm access.
Yes. LynxKite 2000:MM is Python-extensible and supports plug-ins, API integrations, and custom nodes, giving engineers the flexibility to integrate bespoke logic, models, or data pipelines. The integrated code editor makes authoring new operations easy.
LynxKite 2000:MM natively supports NVIDIA GPU clusters for high-speed computation, with automatic fallback to CPU for flexibility. It integrates with NVIDIA RAPIDS (cuGraph) to power 100+ GPU-accelerated graph algorithms and automatically starts and stops GPU microservices to reduce Kubernetes cluster waste.
Yes. LynxKite 2000:MM seamlessly integrates with NVIDIA BioNeMo, allowing you to run pre-trained generative AI models that understand molecular representations like SMILES. These models can be further refined using RDKit directly within the platform.
For Pharma
Yes. LynxKite 2000:MM is designed so that scientists and researchers can work without being programmers. You can create AI workflows in a visual interface, and the agentic assistant helps you build new operations by describing what you need in plain language. When you want more control, everything can be extended in Python.
LynxKite 2000:MM supports omics data (genomics, proteomics), scientific publications, clinical trial data, and molecular/chemical structure data. It integrates diverse sources into a single biomedical knowledge graph for unified analysis.
LynxKite powers Biophytis' AI-driven longevity drug-discovery platform (the MASSIVE project), showcased at NVIDIA GTC 2026 with Nebius. Pharma teams have achieved faster compound screening and reduced late-stage failure risk.
AI automates parts of the discovery process, identifies hidden patterns in biomedical knowledge graphs, and enables in-silico experimentation — helping researchers move faster and more confidently from target identification to preclinical candidates.
Get started with LynxKite
Run the demo in your browser, or co-develop a programme with us.