train models and read papers in one place.
fine-tune deepseek on tinker gpus, run evaluations, and browse research papers — all orchestrated by your agents.
delegate literature reviews, train models on cloud gpus, design experiments, and generate publication-ready latex — ai agents that do the work for you.
npm i -g @synsci/cli
switch between modes mid-session. each one carries a different set of skills and follows a different research workflow.
end-to-end ML research: hypothesis, literature search (pubmed/biorxiv/arxiv/openalex), experimental design, GPU provisioning, training, evaluation, latex paper generation.
builds specialized LLMs from production data. Explores code, designs fine-tunes on Tinker, evaluates, and deploys compounding improvements.
wet-lab + computational biology agent. State-of-the-art on BixBench MCQs with 0% refusal rate. Protein design, genomics, drug discovery, pathway analysis (KEGG/Reactome).
publication-ready LaTeX from scratch. Verified citations, BibTeX, graphical abstracts, figures, camera-ready formatting.
general-purpose coding agent for non-research tasks. Multi-step implementation, file editing, git workflows, debugging.
read-only codebase exploration. Maps architecture, traces dependencies, designs implementation strategies before writing code.
fine-tune deepseek on tinker gpus, run evaluations, and browse research papers — all orchestrated by your agents.
run long-running agents in the browser. kick off a training run or literature review and close your laptop. your agents keep working while you sleep.
every agent runs in an isolated, stateful environment. pause your work and resume exactly where you left off, anytime.
from a lightweight 4-core instance to an 8x H100 cluster, provision exactly the compute your agent needs.
launch evaluations, training runs, or data pipelines instantly from the CLI or dashboard.
kick off a research session in the browser, close your laptop — come back to a finished analysis with outputs, figures, and a literature review.
connect once on the dashboard. credentials sync to every agent, every session.
connect once on the dashboard. your github, huggingface, weights & biases, and other api keys automatically sync to every agent, every session.
provision computing clusters across multiple providers seamlessly.
start researching immediately. everything from package dependencies to python environments is fully managed by our sandboxed runtime.
credits cover all included models. connect your own keys for external services.
all plans include all 6 modes · credential sync · gpu orchestration
