Work starts where people already are.
Messages, mail, meetings, documents, source, and build logs.
Customer-owned inference / workplace-native
Workplace AI, inside your perimeter.
Bring AI into Slack, Teams, email, calendars, WhatsApp, Word, PowerPoint, IDEs, and CI. Your gateway chooses the model. Your infrastructure handles the work. Ovrion is not a hosted inference hop.
Ovrion ships as a customer-run image. Workplace integrations require a subscription. Cloud GPU capacity is optional, pre-funded, and metered by the second.
TRUST BOUNDARY / 01
customer gateway
WORKPLACE REQUEST
SELECTED MODEL
The request path
Trust is a route you can inspect.
A workplace request reaches the gateway you operate, runs on a model endpoint you select, and returns with provenance. Billing and egress evidence remain with the deployment.
Your gateway authenticates and routes.
Signatures, OAuth, API keys, model aliases, billing gates, and provenance.
Your model endpoint does the work.
Local hardware, a private network endpoint, or an explicitly started GPU pod.
Safety, named precisely
Different safeguards deserve different words.
Ovrion does not turn code-level choices into OAuth-level promises. The product names where a boundary is externally enforced and where it is enforced by implementation.
Outlook mailDraft and summarizeMail.Send excluded from OAuth scopeENFORCED
GmailDraft and summarizeGoogle scopes also permit sendCODE-LEVEL
CalendarTentative RSVPNo organizer notification requestedENFORCED
WhatsAppOutbound replyCloud API has no draft responseEXPLICIT SEND
Outlook can structurally exclude mail sending at consent. Gmail cannot. Ovrion's Gmail client contains no send method, but Google grants broader capability to the authorized service account. WhatsApp replies require an outbound Meta API call by design.
Workplace surface
One boundary, many places to work.
Each surface maps to the same customer-run gateway, with the authentication and reply behavior appropriate to that platform.
Slack
Mentions, DMs, thread replies, slash commands, and interactive actions.
V0 HMAC / BOT LOOP GUARD ShippedMicrosoft Teams
Plain replies, Adaptive Cards, bot-added greetings, and mention stripping.
BOT FRAMEWORK JWT / AUDIENCE CHECK ShippedOutlook
Draft replies, thread summaries, and silent tentative RSVP notes.
DELEGATED OAUTH / NO MAIL.SEND ShippedGmail + Calendar
Draft replies, summaries, and tentative RSVP notes through Workspace delegation.
ADMIN DELEGATION / CODE-LEVEL NO-SEND Release candidateVerified Business Cloud webhooks and real outbound text replies.
X-HUB-SIGNATURE-256 / META SEND ShippedWord + PowerPoint
A live task pane that asks the gateway and inserts only when the user chooses.
OFFICE.JS / USER-CONTROLLED INSERT Live add-inGPU economics
Know the rate before capacity starts.
Ovrion sizes a Hugging Face model with its context window, quotes the required GPU tier, checks prepaid runway, and stops the pod before a wallet can go negative. Billing is by warm second, not by token.
QUOTE / CONTEXT-AWAREBEFORE BOOT
Your selected model
Model weightsMeasured
KV cacheIncluded
GPU tierRight-sized
Billed rateVisible first
AUTO-STOP AT ZEROAPPEND-ONLY LEDGER
Commercial model
The engine is open. The workplace layer is licensed.
Run the core cluster freely. Subscribe when you connect workplace surfaces. Buy GPU credit only when customer hardware is not enough.
Core
$0Open-source cluster, OpenAI-compatible gateway, local and private model endpoints.
Pro
$29 / monthWorkplace integration access for one customer-run gateway.
Business
$99 / monthBusiness subscription record and priority-support tier. Final entitlement packaging remains subject to the published license terms.
GPU credit
PrepaidRunPod cost plus the configured fee, debited per warm second and never below the cost floor.
Deploy on customer hardware
Put useful AI where your trust boundary already is.
Pull the published image, connect a model endpoint, configure the workplace surfaces you need, and verify the route yourself.