Skip to main content
A project is the working unit of Galagos. Behind the scenes, it is a persistent, isolated Linux sandbox with its own filesystem, installed packages, and chat history. The agent runs real bash, R, and Python inside it; you watch from the chat interface. When the analysis is over, every command, every script, every output is still there — months later, in the same project — so the work stays reproducible.

What’s inside a project

Open a project and the layout falls into three regions:
  • Chat input at the bottom — where you type your research question.
  • File system panel on the right — every file you upload, every file the agent produces, with size and timestamp. Drag-and-drop adds files; click to view them in the right viewer.
  • Status bar at the top — shows whether the project is Ready or Working, the budget remaining, and (when relevant) a coloured strip with any background jobs.
A wrench icon on the right opens the project’s Skills panel. A reasoning pane appears alongside the chat when the agent has a plan to show you.

Creating a project

1

Open the Projects dashboard

Log in to Galagos and click Projects in the sidebar. You’ll see every project in the current workspace, with last activity and chat count on each card.
2

Create a new project

Click New project in the top-right corner. Give it a descriptive name — for example, “RNA-seq cohort A, batch 3”.
3

Upload your files

Drag and drop files onto the file panel, or click Upload. The platform accepts FASTQ, BAM, VCF, H5AD, CSV, PDB, FASTA, Newick, BigWig, and many other formats.
4

Start your analysis

Type your question into the chat. The agent writes its plan in the reasoning pane before compute starts.

How the sandbox runs

The sandbox is not a chatbot wrapper. It is a real Linux environment that:
  • Runs bash, R, and Python natively.
  • Installs packages on demand (pip, install.packages, apt) and keeps them persisted across sessions.
  • Downloads datasets server-side — your laptop does not stay open while a 30 GB SRA accession arrives.
  • Shows background jobs in an orange strip above the chat input with the running process name. The user-facing message: “1 background process: wget”.
  • Continues running while your browser is closed. Sessions are picked up exactly where you left them when you return.
If a long job needs your attention or finishes overnight, Galagos emails you. You do not need to keep the tab open.

The step inspector

Click any step in a run and the inspector reveals the exact command, working directory, stdout, and stderr. This is the level of transparency that makes a result defendable. There is no hidden glue — every command is real, recorded, and re-readable.

Storage and the Files panel

Each project has its own storage quota. The limit applies per project, not per account. The file panel shows used and total storage when you hover the storage line (for example, “5.4 MB of 4.0 GB”). No silent quota failures — if you are running tight, the agent surfaces it before a write fails.
Intermediate files (sorted BAMs, partial assemblies) are often the largest consumers. Delete them after downstream analysis succeeds — the agent can always re-generate them from the originals.

Active project limits

Your plan also sets how many projects you can keep active. Archiving a project frees a slot without deleting any of its files or results.

Session persistence

Galagos saves project state continuously. Close the browser, switch laptops, come back tomorrow — the chat history, the files, the partial outputs, and any in-flight background jobs are all preserved. See Session persistence for the full behaviour.

Reproducibility and audit

Every action inside a project is recorded:
  • The plan the agent wrote before running.
  • Every shell command, with stdout and stderr.
  • Every package install, with version.
  • Every Skill it consulted.
  • The Limitations paragraph it wrote about its own output.
You can export the entire project — chat history, inputs, agent-written scripts, outputs, and the package-version record — as a single archive. See Reproducibility and audit.

Project isolation

Every project runs in its own sandbox. Files, processes, and installed packages from one project cannot access or interfere with another, even within the same workspace. A misconfigured analysis cannot corrupt data in another project; you can run conflicting tool versions side by side; your work never mixes with anyone else’s.
Don’t paste credentials, API keys, or passwords into uploaded files or chat messages. If an analysis needs external credentials, contact support for the right way to provide them.

Sharing a project

You can share a project with named teammates (read or write access) from within the project. Sharing is per-project — separate from workspace membership — so you can keep most of your work private while opening one project to a collaborator. The audit log records every share.

Workspaces

How projects fit alongside the library, members, and Skills at the workspace level.

Uploading files

Supported formats and best practices for getting data in.

Running an analysis

The conversation loop the agent follows inside every project.

Credits and storage

How usage is metered.