synsc. It runs alongside your shell and gives you a single agent that already knows how to do science. Pick a research mode and the agent loads domain-aware tools, cloud credentials, and the right backend automatically.
@synsci/cli. The dashboard lives at cli.syntheticsciences.ai.
What you get
- Research modes for general research, physics, chemistry, biology, and the model flywheel for building specialized LLMs from production data. Each mode primes the agent with domain conventions and routes calls to the right backend.
- Bundled domain tools for each mode. From DFT and molecular dynamics in chemistry to GRPO and vLLM in flywheel. The agent loads them automatically when your prompt calls for them.
- Credential sync for Hugging Face, Weights & Biases, Modal, Lambda, AWS, and GCP. Tokens come from your dashboard and reach subprocesses without ever touching disk.
- Included credits that cover the agent’s work. Pre-call balance checks block runaway loops before they bill.
Two agents
buildis the default. Full filesystem and shell access for development work.planis read-only. Same context and tools, but cannot mutate anything.
@general in any message to delegate a complex search or multi-step task to a sub-agent that runs in its own context.
Where it fits
| You want to | Reach for |
|---|---|
| Run scientific compute work in your terminal | Synsci CLI with the right research mode |
| Track hypotheses, experiments, and decisions over time | Thesis |
| Use both, with the agent writing code and Thesis storing what was tried | Run both, paired through MCP |
What’s next
- Install Synsci CLI.
npx synscidoes the entire setup. - Quickstart. Connect, pick a mode, send your first prompt.
- Web UI. The recommended day-to-day interface.
- Research modes. What each mode covers and how to switch.