in the browser
opencode
A full AI coding agent with a web UI, one click from the instance list. It opens inside your project, reads and edits files, runs shell commands, and executes code in the same Python environment as your notebooks.
agent sandbox
Let your agents run loose. On their own machine.
Give your coding agents a machine of their own. opencode runs in the browser, Codex and Claude Code in the terminal, all on a dedicated cloud VM with Python and R ready to go. Whatever they plot is on the record.
Models served with no API keys to configure. Every plan includes a compute allowance.
youRefit the survival model with stage as a covariate and regenerate the KM plot.
agentReading examples/tcga_lung_classifier.ipynb…
$ python -c "import lifelines; …"
Added stage to the Cox model. Updated plot_km() and re-ran the notebook.
gofigr: published km_by_stage.png → rev 15 · code, data, env captured
Done. Stage is a significant covariate in the refit; the updated curves are in your workspace.
trusted by top organizations
what’s in the box
Three agents, one machine, the same files
in the terminal
Codex CLI
OpenAI’s Codex, pre-wired to GoFigr’s model gateway. Open a terminal from JupyterLab or code-server and run it. No API key to paste, nothing to configure.
in the terminal
Claude Code
Anthropic’s Claude Code, installed and ready. Sign in with your own Claude subscription or API key and it works on the same files as everything else. Add the GoFigr plugin and it knows how to publish figures with provenance.
how it works
From prompt to reviewed figure
- 1
Launch
Pick a machine size and launch. In a few minutes you have a dedicated cloud VM with Python, R, JupyterLab, code-server, and the agents installed.
- 2
Hand it the task
Open opencode from the instance list, or run codex or claude in a terminal. Point the agent at your notebooks and data and describe what you want.
- 3
Let it run
The agent edits, executes, and iterates on its own machine. If it breaks something, it breaks its own box. Your laptop and your team’s environments are untouched.
- 4
Review the figures
Every figure the agent produces lands in GoFigr with the code, data, and environment behind it. Review the work like a pull request, months later if you need to.
why not your laptop
Built for agents that run unattended
Isolated by construction
Each sandbox is its own virtual machine. Agents get a real shell with real permissions, and the blast radius ends at the machine boundary.
It’s already a data science box
Python, R, pandas, scikit-learn, JupyterLab, code-server, and RStudio Server are preinstalled. The agent works in the same environment you would.
Persistent, not disposable
Files, installed packages, and agent sessions live on a data volume that survives stop and start. Long analyses pick up where they left off.
Models without keys
The instance authenticates to GoFigr’s model gateway on its own. Usage is metered against your plan’s AI allowance and shown in the app.
Auto-stop when idle
Sandboxes park themselves after a period of inactivity, so an agent you forgot about doesn’t become a bill you didn’t expect.
Provenance on everything
This is the part no generic sandbox has. Every figure an agent makes links back to its exact code, data, and environment.
models, no keys
Pick the model, skip the key management
opencode and Codex talk to GoFigr’s model gateway out of the box. Switch between frontier and open-weight models per task; usage is metered against your plan’s AI allowance and broken down by model in the app.
- Claude Sonnet 5opencode default
- Claude Opus 5
- Claude Haiku 4.5
- GPT-6 Astra
- GPT-5.6 Sol / Terra / LunaTerra: Codex default
- GPT-OSS 120B
- Kimi K2.5
- DeepSeek V3.2
- Qwen3 Coder Next
- GLM-5
- Devstral 2
- MiniMax M2.5
the fine print
Good to know
Is Claude Code included?
Claude Code is installed, but it runs on your own Anthropic account or API key. Anthropic’s terms don’t allow us to resell Claude Code usage, so it isn’t metered through GoFigr. opencode and Codex use GoFigr’s gateway and need no keys at all.
What does it cost?
Compute bills per minute while the sandbox is running, storage bills for the data volume, and model usage draws on your plan’s AI allowance. Every plan includes a compute allowance; a card on file is required to launch, and it’s only charged for usage beyond that allowance.
Can the agent reach my data?
Anything on the sandbox’s data volume, yes: that’s the point. Upload files, clone a repository, or download from your data sources, and the agent works on them there. Your local machine is never involved.
What happens when I stop the sandbox?
Compute billing stops. Files, installed packages, and agent sessions are preserved on the data volume and come back on the next start. Deleting the sandbox removes the volume.
Can I bring my own model keys?
Yes. Each agent’s configuration is an ordinary file on your data volume. Point it at your own provider and it stays that way across restarts.
give your agents a home
A sandbox in minutes. A paper trail forever.
Simple hourly pricing with a compute allowance on every plan — view pricing. Want the agents inside your own network? See GoFigr for enterprise.