The open-source AI workbench for scientific research

Give it a goal. It reads the literature, writes and runs code, runs the experiments, and writes up what it found.

Install · Quickstart · Docs · Atlas

---

OpenScience is an AI workbench for scientific research. You give it a goal, and it works through the research loop the way a capable collaborator would. It reads the papers that matter, forms a hypothesis, writes and runs code, runs experiments on real compute, queries the major scientific databases, and writes up the result. It runs as a workspace in your browser and works with any frontier or open-weight model from Anthropic, OpenAI, Google, and dozens of other providers, using your own API keys. No account is required.

It is model-agnostic, open source, and built to do real work in machine learning, biology, physics, and chemistry.

What it does

• Runs the whole loop. Literature review, hypothesis, code, experiment, analysis, and write-up, in one continuous session.
• One adaptive Research agent. A single user-facing collaborator handles the task end to end, loads domain skills when useful, and can delegate bounded Explore, Execute, or Review work internally. Normal and Ultra efforts control how widely it investigates; plan mode stays read-only.
• 295 bundled skills. Training (DeepSpeed, PEFT, TRL), evaluation, dataset work, molecular and clinical biology, cheminformatics, papers and LaTeX, figures, and cloud compute (Modal, Tinker, and others).
• Scientific databases as tools. UniProt, PDB, Ensembl, ChEMBL, PubChem, arXiv, OpenAlex, Semantic Scholar, and around 30 more, queryable directly by the agent.
• A real workspace. A browser UI with a file tree, an editor, a terminal, session history, and inline rendering for molecules, structures, genomes, and plots.
• Extensible. LSP integration, MCP servers, plugins, custom agents and commands, and a TypeScript SDK.

Install

Install with npm, then open the workspace:

npm install -g @synsci/openscience
openscience

The command is openscience, and it opens the workspace in your browser without requiring an account or setup wizard. Connect ChatGPT/Codex, your own provider keys, or Atlas managed models from Customize → Models when you want them. If you would rather not install it globally, npx synsci does the same thing in a single step:

npx synsci

Platform binaries are also attached to GitHub Releases; see the changelog for what's new in each version.

Linux installs require kernel 5.1 or newer. Glibc builds require glibc 2.17 or newer, and musl builds are published separately. CentOS 7's stock 3.10 kernel is not supported even though its glibc version meets the minimum; use a newer host kernel or VM.

Quickstart

Set an API key from any provider (ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, and so on) and start the workspace:

export ANTHROPIC_API_KEY=sk-ant-...
openscience

openscience opens the workspace in your browser. Your keys stay on your machine and requests go straight to the provider. You can also run openscience keys add to store a key from the terminal, add keys from the Credentials panel, and pick a model from the model selector. To open the workspace in a specific project:

openscience ~/code/my-project

Atlas

Atlas is Synthetic Sciences' managed platform. It gives you a curated set of frontier models billed from a prepaid wallet, so you do not need per-provider keys, plus a persistent research graph and cloud compute. OpenScience works with Atlas but never requires it.

openscience login          # connect your Atlas account
openscience wallet         # check your balance and top up

Bring-your-own-key usage is always free and is never gated — Atlas only meters the models it serves. Use openscience status to see what you are connected to, and openscience logout to disconnect.

How it works

OpenScience runs a local server that hosts the workspace UI, the agent runtime, the complete default skill library, and the tool layer. The agent plans with a research harness, calls tools (shell, editor, LSP, MCP servers, scientific connectors, and skills), and streams its work back to the browser. Models are routed per request, so you can switch between providers or run local models without changing anything else. Sessions, skills, artifacts, and provenance are stored on disk. Atlas adds optional managed models, credential sync, research graphs, library search, and cloud publishing after login.

Path	Contents
backend/cli	The CLI, server, provider integrations, sessions, and skills
frontend/workspace	The browser workspace UI, served by the CLI
frontend/docs	The documentation and session-share site
tooling/sdk/js	The TypeScript SDK
tooling/plugin	The plugin runtime

Configuration

Global config lives in ~/.config/openscience/openscience.json. Project config lives in openscience.json or a .openscience/ directory at the repo root (schema). Custom agents, commands, tools, plugins, and themes load from those directories.

Development

You need Bun 1.3 or newer.

bun install
bun dev
bun run typecheck
bun run --cwd backend/cli test
bun run --cwd backend/cli build

bun dev runs the workspace from source, and bun run --cwd backend/cli build produces the platform binaries.

See ARCHITECTURE.md for how the system fits together, CONTRIBUTING.md for how to contribute, AGENTS.md for the style guide, and CODE_OF_CONDUCT.md for community standards.

Security

The permission system keeps you aware of what the agent is doing; it is not an isolation boundary by itself. OpenScience enables its OS execution sandbox by default: macOS Seatbelt or Linux bubblewrap confines commands to the workspace and approved paths and denies network egress. Routine terminals, kernels, shell commands, and local jobs can run immediately inside that verified boundary; remote jobs, kernel environment changes, project extensions, and host execution still require explicit project trust. The default fails closed when no backend is available, and the sandbox is not a full jail, so run inside a container or VM for hostile code. Managed Atlas tokens stay out of general subprocess environments, arbitrary Python/R kernels receive a minimal environment, and credential-shaped values are redacted from output. To inspect or verify containment, run openscience sandbox and openscience sandbox test; to report a vulnerability, see SECURITY.md.

License

Apache License 2.0. See LICENSE and NOTICE.

OpenScience is an independent project. It is not affiliated with, endorsed by, or sponsored by Anthropic. "Claude" is a trademark of Anthropic, PBC, used here only to describe compatibility.