Nygen's hosted CyteOnto service compares cell-type annotations against the Cell Ontology. Its default language and embedding providers are Nebius, using Kimi K2.6 and Qwen3-Embedding-8B. Users can run comparisons without supplying API keys because the hosted service supplies its own provider keys.
CyteOnto uses a language model to split compound labels and write descriptions. An embedding model embeds those descriptions for matching against ontology terms and scoring annotation agreement. Its documentation identifies Nebius Token Factory endpoints. The hosted service supports alternative providers and falls back to Fireworks for descriptions and OpenRouter for embeddings when primary requests fail.
Descriptions and embeddings are cached per model. On the first embedding failover, CyteOnto embeds 2,902 ontology descriptions with OpenRouter and saves them. It re-embeds labels from the affected call so its scores use one embedding model. The tutorial records 18 Nebius language-model requests totaling 10,721 tokens, comprising 8,859 input and 1,862 output tokens.
CyteOnto defaults to Nebius Token Factory for language-model descriptions and embeddings. When embedding requests fail over to OpenRouter, it re-embeds labels from the affected call so all scores use one embedding model. The first failover creates fallback embeddings for 2,902 ontology descriptions. The hosted API disables the optional PubMed tool by default.
CyteOnto models and providersListed 2 Nov 2026 · GitHub · Web pageCyteOnto defaults to Nebius Token Factory for language-model descriptions and embeddings. When embedding requests fail over to OpenRouter, it re-embeds labels from the affected call so all scores use one embedding model. The first failover creates fallback embeddings for 2,902 ontology descriptions. The hosted API disables the optional PubMed tool by default.
CyteOnto hosted API reference
Listed 2 Nov 2026 · GitHub · Web page
Nygen's hosted CyteOnto API defaults to Nebius for language and embedding models. Failed descriptions fall back to Kimi K2.6 on Fireworks, while failed embeddings fall back to Qwen3-Embedding-8B on OpenRouter. The first embedding failover also embeds 2,902 ontology descriptions, making that job take longer.
CyteOnto hosted API referenceListed 2 Nov 2026 · GitHub · Web pageNygen's hosted CyteOnto API defaults to Nebius for language and embedding models. Failed descriptions fall back to Kimi K2.6 on Fireworks, while failed embeddings fall back to Qwen3-Embedding-8B on OpenRouter. The first embedding failover also embeds 2,902 ontology descriptions, making that job take longer.
CyteOnto executed tutorial notebook
Listed 2 Nov 2026 · GitHub · Web page
CyteOnto's recorded tutorial uses Nebius for language-model descriptions and embeddings. Its language-model usage records 18 requests and 10,721 tokens, comprising 8,859 input and 1,862 output tokens.
CyteOnto executed tutorial notebookListed 2 Nov 2026 · GitHub · Web pageCyteOnto's recorded tutorial uses Nebius for language-model descriptions and embeddings. Its language-model usage records 18 requests and 10,721 tokens, comprising 8,859 input and 1,862 output tokens.
CyteOnto hosted API quickstart
Listed 2 Nov 2026 · GitHub · Web page
Nygen's hosted CyteOnto quickstart illustrates Nebius as the language and embedding provider. The service sends submitted labels to its model providers and retains each job's results, descriptions and embeddings.
CyteOnto hosted API quickstartListed 2 Nov 2026 · GitHub · Web pageNygen's hosted CyteOnto quickstart illustrates Nebius as the language and embedding provider. The service sends submitted labels to its model providers and retains each job's results, descriptions and embeddings.
Original source
Listed 3 Nov 2026 · raw.githubusercontent.com · Web page
Nygen's hosted CyteOnto deployment defaults to Nebius for language and embedding models and supplies provider keys when requests omit them. Descriptions and embeddings are cached per model. The first embedding failover to OpenRouter embeds 2,902 ontology descriptions and saves the artifact on a Modal volume, delaying that run.
Original sourceListed 3 Nov 2026 · raw.githubusercontent.com · Web pageNygen's hosted CyteOnto deployment defaults to Nebius for language and embedding models and supplies provider keys when requests omit them. Descriptions and embeddings are cached per model. The first embedding failover to OpenRouter embeds 2,902 ontology descriptions and saves the artifact on a Modal volume, delaying that run.
Nygen develops software for analyzing and interpreting single-cell omics data, serving academic researchers, biotechnology companies and pharmaceutical teams. Operated by Nygen Analytics AB in Lund, Sweden, its main products are ScarfWeb and CyteType. ScarfWeb provides a browser workspace for quality control, clustering, visualization, statistical analysis and collaboration without coding. CyteType uses AI to identify cell types and functional states, providing ontology labels, marker evidence and confidence scores. Researchers use these tools to study disease biology, investigate drug targets and compare cellular responses across experiments.
Nygen develops single-cell analysis and cell annotation tools for biological research and drug discovery. · Health
How they use Nebius Token Factory
Nygen's hosted CyteOnto service compares cell-type annotations against the Cell Ontology. Its default language and embedding providers are Nebius, using Kimi K2.6 and Qwen3-Embedding-8B. Users can run comparisons without supplying API keys because the hosted service supplies its own provider keys.
CyteOnto uses a language model to split compound labels and write descriptions. An embedding model embeds those descriptions for matching against ontology terms and scoring annotation agreement. Its documentation identifies Nebius Token Factory endpoints. The hosted service supports alternative providers and falls back to Fireworks for descriptions and OpenRouter for embeddings when primary requests fail.
Descriptions and embeddings are cached per model. On the first embedding failover, CyteOnto embeds 2,902 ontology descriptions with OpenRouter and saves them. It re-embeds labels from the affected call so its scores use one embedding model. The tutorial records 18 Nebius language-model requests totaling 10,721 tokens, comprising 8,859 input and 1,862 output tokens.
CyteOnto defaults to Nebius Token Factory for language-model descriptions and embeddings. When embedding requests fail over to OpenRouter, it re-embeds labels from the affected call so all scores use one embedding model. The first failover creates fallback embeddings for 2,902 ontology descriptions. The hosted API disables the optional PubMed tool by default.
CyteOnto models and providersListed 2 Nov 2026 · GitHub · Web pageCyteOnto defaults to Nebius Token Factory for language-model descriptions and embeddings. When embedding requests fail over to OpenRouter, it re-embeds labels from the affected call so all scores use one embedding model. The first failover creates fallback embeddings for 2,902 ontology descriptions. The hosted API disables the optional PubMed tool by default.
CyteOnto hosted API reference
Listed 2 Nov 2026 · GitHub · Web page
Nygen's hosted CyteOnto API defaults to Nebius for language and embedding models. Failed descriptions fall back to Kimi K2.6 on Fireworks, while failed embeddings fall back to Qwen3-Embedding-8B on OpenRouter. The first embedding failover also embeds 2,902 ontology descriptions, making that job take longer.
CyteOnto hosted API referenceListed 2 Nov 2026 · GitHub · Web pageNygen's hosted CyteOnto API defaults to Nebius for language and embedding models. Failed descriptions fall back to Kimi K2.6 on Fireworks, while failed embeddings fall back to Qwen3-Embedding-8B on OpenRouter. The first embedding failover also embeds 2,902 ontology descriptions, making that job take longer.
CyteOnto executed tutorial notebook
Listed 2 Nov 2026 · GitHub · Web page
CyteOnto's recorded tutorial uses Nebius for language-model descriptions and embeddings. Its language-model usage records 18 requests and 10,721 tokens, comprising 8,859 input and 1,862 output tokens.
CyteOnto executed tutorial notebookListed 2 Nov 2026 · GitHub · Web pageCyteOnto's recorded tutorial uses Nebius for language-model descriptions and embeddings. Its language-model usage records 18 requests and 10,721 tokens, comprising 8,859 input and 1,862 output tokens.
CyteOnto hosted API quickstart
Listed 2 Nov 2026 · GitHub · Web page
Nygen's hosted CyteOnto quickstart illustrates Nebius as the language and embedding provider. The service sends submitted labels to its model providers and retains each job's results, descriptions and embeddings.
CyteOnto hosted API quickstartListed 2 Nov 2026 · GitHub · Web pageNygen's hosted CyteOnto quickstart illustrates Nebius as the language and embedding provider. The service sends submitted labels to its model providers and retains each job's results, descriptions and embeddings.
Original source
Listed 3 Nov 2026 · raw.githubusercontent.com · Web page
Nygen's hosted CyteOnto deployment defaults to Nebius for language and embedding models and supplies provider keys when requests omit them. Descriptions and embeddings are cached per model. The first embedding failover to OpenRouter embeds 2,902 ontology descriptions and saves the artifact on a Modal volume, delaying that run.
Original sourceListed 3 Nov 2026 · raw.githubusercontent.com · Web pageNygen's hosted CyteOnto deployment defaults to Nebius for language and embedding models and supplies provider keys when requests omit them. Descriptions and embeddings are cached per model. The first embedding failover to OpenRouter embeds 2,902 ontology descriptions and saves the artifact on a Modal volume, delaying that run.
Nygen develops software for analyzing and interpreting single-cell omics data, serving academic researchers, biotechnology companies and pharmaceutical teams. Operated by Nygen Analytics AB in Lund, Sweden, its main products are ScarfWeb and CyteType. ScarfWeb provides a browser workspace for quality control, clustering, visualization, statistical analysis and collaboration without coding. CyteType uses AI to identify cell types and functional states, providing ontology labels, marker evidence and confidence scores. Researchers use these tools to study disease biology, investigate drug targets and compare cellular responses across experiments.