Questions, answered.

Normally your AI reads the file as text — turning your spreadsheet into an approximation. Numbers get fuzzy, formulas disappear, and you get its best guess. With Sidecar the file stays a file: your AI opens it in a real workspace, runs real analysis, and shows you the actual results. You never see the mechanics — just the output.

Ask your AI to read a spreadsheet directly and it answers from what it “sees” in the text — the numbers are approximate and the analysis isn’t really running. With Sidecar your AI hands the real work to a real workspace: the spreadsheet opens properly, the statistics run for real, the chart is drawn from actual data. The answer is what the data says, not what the model thinks it probably says.

Yes. Every Sidecar workspace runs in complete isolation — your data never touches anyone else’s. Your files are only present while the workspace is active, and when it ends the workspace is destroyed. Nothing is stored unless you explicitly choose to save it.

No. Your AI handles everything. You ask questions, you see results, and you never need to understand or touch anything technical.

Yes. Workspaces can be restarted — your AI remembers what it did, the data reloads, and you continue exactly where you were. It’s like reopening a document with everything as you left it.

You buy credit — 1 credit is $1. While your workspace is working, credit burns at an hourly rate based on its size, from $0.25/hr for everyday work up to $0.50/hr for the largest. Pro and Enterprise add a subscription that unlocks every flavor and higher limits. Set a monthly spend cap and you’ll never exceed your budget, and when your balance reaches zero workspaces stop on their own. No surprise bills.

Every workspace is its own container, created for one person and one piece of work, and destroyed with it. Workspaces share no memory, no filesystem, and no network path to one another. There is no pooled compute that two customers’ work passes through.

Inside that workspace’s own storage, and nowhere else. It is never written to shared infrastructure. When the workspace ends, it and its storage are destroyed together — what survives is only what you explicitly chose to save.

Credentials for databases, APIs and third-party tools are encrypted at rest with symmetric encryption, and the key lives in environment configuration rather than in the database beside the data it protects. The workspace itself never receives them: requests leave through a boundary that attaches the credential on the way out, so a credential cannot be read out of the workspace, printed by accident, or appear in a log, in the viewer, or in any output you can see.

Outbound traffic passes through a boundary we control. Private address ranges, loopback, link-local and cloud metadata endpoints are refused there, so a workspace cannot reach internal services or anything on a private network. Public sources your analysis needs stay reachable — the policy is a denylist, because an allowlist is not workable for open-ended analysis.

Sidecars runs on Cloudflare. A Worker handles sign-in, workspace lifecycle and routing; each workspace is a container owned by its own Durable Object, so a workspace’s lifetime is that object’s lifetime and there is no separate fleet to fall out of step with it. Each workspace runs the flavor you chose, on its own compute.

Enterprise plans can include dedicated deployment. Contact us for details.

We’re establishing our compliance posture during the private beta and will document it honestly as it’s in place. Reach out and we’ll share exactly where things stand today.