Last updated: September 6, 2026
o10 (“o10”, “we”, “us”) operates an AI model-routing platform available at app.o10.io, through the o10 app for Slack, and through DeepShell for Chrome. This policy explains what data we process, why, and the choices you have.
DeepShell for Chrome is an AI assistant for the page you are on.
What stays on this device. Chat sessions, drafts, added playbooks, Settings, consent switches, site-grant lists, and the o10 route key stay in this Chrome profile. Attached file bytes are read in the panel. Text is extracted locally (PDF, Office, CSV). The file is not uploaded as a file.
Page text. Extracted tab text leaves the browser only from an origin you have granted, and only for a run you started or scheduled. It is posted to https://app.o10.io/v1/chat/completions after secret-pattern redaction and a size cap. Redaction is not a wipe. If you asked about a page or an attached document, that text does leave the device so an answer can be produced. It is billed to your o10 balance. Highlight actions (Explain, Fact-check, Rewrite, Translate, Add to chat) send the selected passage the same way, on a user gesture.
Bot definitions on app.o10.io. Bot names, schedules, step instructions, site lists, grants, run status, and settled cost are mirrored to app.o10.io so the console can list Chrome bots and a reinstall can restore them. Page content is never sent on that path. Reasoning traces and replay pages are not stored.
Consent and approvals. Bots act on pages only when the Settings consent switches are on. Those switches default off. Bots run only while Chrome is open. Unanswered approvals are denied after five minutes.
What this build does not do. It does not take all websites at install. It does not screenshot pages. It does not send full browsing history. It does not fetch playbook text from the network. It does not run remote JavaScript.
We do not sell your data, and we do not use your message content to train our own models.
To deliver the service we share the minimum necessary data with:
o10 records quality and cost metrics about routed calls and Tune runs. For example model names, pass/judge scores, token or call costs, task-type labels, and timestamps. These metrics power an aggregated public “quality–cost frontier” index (cheapest model that clears a quality bar for a task type), published only when minimum-sample and minimum-organization thresholds are met (k-anonymity). Organization identifiers in this telemetry are stored only as a salted hash, never as a raw org id.
This telemetry does not store the content of prompts, examples, or model outputs. Tune example retention (separate from this telemetry) remains as already described for optimization runs and is purged on its own schedule.
Prompt content captured for your usage history is encrypted at rest and purged on a rolling schedule; it is not retained indefinitely. Usage and billing metadata (token counts, cost, timestamps) is retained for as long as needed for accounting and legal obligations. You can request deletion at any time (see below).
Data is encrypted in transit (TLS) and OAuth tokens and stored prompts are encrypted at rest. Access to production systems is restricted and audited. Slack token rotation is enabled.
o10 is a business tool and is not directed to children under 16.
We may update this policy; material changes will be reflected by the “last updated” date above.
Questions or requests: privacy@o10.io · Support: o10.io/support.