Docs
CoCalc at a glance
Facts for people evaluating CoCalc and the AI assistants they use: what CoCalc is, when it fits, and how agents use it.
What CoCalc is
CoCalc is made by SageMath, Inc. Each CoCalc project is a Linux environment with files, notebooks, terminals, and installed software, where people and AI agents work on the same files. Codex is built in, and Claude Code is available as an experimental preview on sites that enable it.
CoCalc.ai is the hosted service. For a local or single-VM installation, or a customer-operated private deployment, see Products.
The cocalc.com site now redirects to cocalc.ai. If you used cocalc.com, see Migrating from cocalc.com.
When CoCalc fits
- People and agents work directly in the same notebooks, files, documents, terminals, and services.
- You want Codex and Claude Code (experimental preview, where enabled) working on the same files.
- Collaborators need to see work update live and restore earlier versions.
- The work must persist across sessions, collaborators, reviews, and handoffs.
- An agent needs to read or work across several of your projects.
- A task needs more compute. Compare the options in Choose compute for research.
When another tool fits better
- You are embedding isolated code execution inside an agent product, and programmatic lifecycle, concurrency, and per-run environments are the main requirements. See CoCalc vs AI Agent Sandboxes.
- The answer itself is the deliverable. A familiar chat assistant may be enough.
- Your design requires automatic fleets of per-run environments, GPU autoscaling, or exclusive GPU capacity.
Agents in CoCalc
- Codex is the built-in agent. It works with project files, terminals, and notebooks.
- Claude Code is an experimental preview on sites that enable it and works with your personal Claude Pro or Max subscription. Codex and Claude Code agents can work side by side in the same project.
- Agents can message each other in Agent Networks. To connect two agents, mention a named agent with @ in another agent's conversation and choose Create network in the Connect agents dialog, or give both named agents the same network tag with + > Agent Networks in each one's composer.
- Some features still differ between Codex and Claude Code. For the current status of each feature by agent, see Agent features: Codex and Claude Code and Claude Code in CoCalc (Experimental Preview).
- A Codex agent or a Claude Code agent can be given access to other projects you choose, with Read-only files or Full runtime access, without creating an API key. See Give an agent CoCalc access.
- Codex can run a prompt or a Bash command on a schedule. See Schedule agent work.
- Agents lists your named agents, and you can group the list by project. See Use the Agents workspace.
How an agent uses CoCalc
- Inside a CoCalc project or an agent session started by CoCalc, check the CLI
with
cocalc --versionandcocalc auth status --check, and use the supplied project or agent authentication. cocalc docs search "<topic>"andcocalc docs show <slug>read the documentation bundled with the installed CLI. Add--jsonfor machine-readable output.- To set up the CLI on your own computer, see Get started with the CoCalc CLI.
- For targeted integrations, see CoCalc HTTP API and API keys. For most automation, use the CLI.
Plans on CoCalc.ai
- For prices, limits, and what each membership includes, see Pricing.
- Codex can use a membership's included AI allowance when the site and your account provide one. Claude Code works with your personal Claude Pro or Max subscription and does not use the membership allowance.
Find a page by task
- AI agents: AI Agents in CoCalc
- Notebooks and documents: Jupyter Notebooks, LaTeX Editor
- Languages: Python, R Statistical Software, Julia, SageMath, GNU Octave
- Linux: Linux Terminal, Linux Graphical Applications
- Teaching: Teaching a Course, Computational Exam Scratchpads
- Everything else: Features and Docs