A group of computational biologists has released CellForge, an open-source system that uses a team of large-language-model agents to design, code and train virtual cell models directly from raw single-cell multi-omics data. According to an accompanying arXiv preprint dated 4 August 2025, the framework breaks the task into specialised agents responsible for dataset profiling, method design and experiment execution, then iteratively reconciles their proposals to generate executable code. Benchmarking on six perturbation-response datasets covering gene knockouts, drug treatments and cytokine stimulation shows CellForge exceeding task-specific state-of-the-art baselines; one test cited by the developers indicates a 22 percent improvement in predictive accuracy. The authors have made both the preprint and the source code available on GitHub, positioning CellForge as a turnkey layer for laboratories that lack extensive software expertise but need bespoke virtual cell models. The release adds to a flurry of open-source bio-AI tools unveiled this week, including MultiNano for single-base m6A RNA modification detection from nanopore reads, SingleRust for high-throughput single-cell data processing in Rust, and Hi-Cformer, a transformer architecture that imputes and compresses single-cell Hi-C contact maps. Collectively, the new toolkits aim to lower the technical barrier to large-scale cellular modelling and analysis.