Security

Your meetings are the most sensitive thing you own.

Onvoi holds transcripts, decisions and internal knowledge, and hands them to software that acts. This page describes the controls around that — and, at the bottom, the ones we do not have yet.

Data

Everything is classified before it is stored.

Each collection carries a classification that sets how it is protected and who may reach it. When a record mixes levels, it takes the higher one.

Critical

Credentials, OAuth tokens and API keys. Encrypted at rest, never written to logs, never returned in an error.

Confidential

Transcripts, chat messages, knowledge facts, execution traces and uploaded files. Encrypted in transit, scoped to their owner and organisation.

Internal

Task metadata, entity names, skills and org structure. Available to authenticated users within scope.

Public

Marketing pages and published documentation. No protection required.

Access

Scope is enforced on the way in, not filtered on the way out.

Authorisation is checked at the route boundary before a handler runs, and again during knowledge retrieval at every hop of the graph.

Stateless sessions

Short-lived access tokens in HTTP-only cookies, with refresh tokens rotated on use. No server-side session store to steal.

Organisation and team scoping

Every scoped route resolves your membership first. Cross-organisation reads are refused at the resolver, not trimmed from a response.

Encrypted third-party tokens

Connected-service credentials are encrypted with AES-256-GCM before they touch the database.

Personal stays personal

Behavioural inferences about an individual are visible only to that individual — administrators included.

Agents

The interesting attack surface is the agent, so that is where the controls are.

An agent that reads untrusted content and holds real credentials is a genuinely new risk. These are the specific mitigations, not a posture statement.

Injection screening before dispatch

Incoming task text is screened by pattern matching first and a model only where the result is ambiguous. Repeated attempts against one account escalate and can suspend dispatch.

Capability constraints at run time

An agent is given an explicit set of permitted tool classes for a task. A step outside that set is rejected as it happens, not noted afterwards.

Confidential content withheld

Before an instruction goes to an outside agent, sensitive sections are stripped. If the task itself is confidential, it does not leave at all.

Hard execution ceilings

Server-side caps on steps, retries, blocked events, runtime and spend apply regardless of how an agent is configured locally.

Fail closed in production

When a security check cannot run, strict mode blocks the work rather than letting it through unchecked.

A stop switch

Administrators can put every agent in the organisation back to zero autonomy, terminate running work and disconnect them, in one action.

Privacy in practice

Personal data is stripped from the places engineers look.

Logs, error reports and internal admin views are the usual route by which customer data leaks sideways. Onvoi masks them at the point of writing rather than trusting reviewers.

  • Detected and masked automatically. Email addresses, phone numbers, national ID numbers, payment card numbers and IP addresses are replaced before an error record or execution trace is stored.
  • Credentials are redacted from logs structurally. Authorisation headers, cookies, tokens and passwords are stripped by the logger itself, not by convention.
  • Retention is yours to set. Choose how long meetings and conversations are kept. Enforcement runs nightly, and can be trialled in dry-run first.
  • Deletion means deletion. Removing a meeting removes its recording, transcript, derived tasks and extracted facts. An erasure request reaches every collection holding that person's data.
  • Your content does not train shared models. Our model provider is under commercial terms that exclude API data from training, and nothing you produce is used to improve another customer's results.
Subprocessors

Who else touches your data.

These are the third parties that may process customer data on our behalf. Each is reassessed annually and whenever its terms change. Optional services appear only if you switch them on.

SubprocessorPurposeData processedLocation
Amazon Web ServicesHosting, storage and content deliveryAll application data at rest and in transitUnited States
MongoDB AtlasPrimary databaseAll application dataUnited States
AnthropicLanguage modelsPrompts containing meeting, chat and knowledge contextUnited States
DeepgramTranscription and speaker separationMeeting audioUnited States
Recall.aiMeeting bot recordingMeeting audio, while a bot is in the callUnited States
Voyage AIText embeddings for searchShort text excerpts from your contentUnited States
StripePayment processingBilling details; no meeting or task contentUnited States

Additional providers may be used for optional features you enable — for example an alternative transcription provider, a cloud browser for hosted execution, or a model provider you choose to bring yourself. Write to security@onvoi.ai for the current full list.

When something goes wrong

We have a plan, and it is written down.

  • Severity is classified on a fixed scale. Data exposure and incorrect external communication sit at the top and trigger the immediate-shutdown path.
  • Alerts escalate on a ladder. An unacknowledged critical alert moves from the on-duty engineer to backup contacts on a timer rather than sitting in an inbox.
  • Serious incidents require a written post-mortem. Cause, impact and mitigation must be recorded before the incident can be closed.
  • Customers are told. If an incident affects your data, we contact you directly with what we know, when we knew it, and what we are doing.
Responsible disclosure

Found something? Please tell us.

Write to security@onvoi.ai with enough detail to reproduce it. We will acknowledge, keep you updated, and credit you if you would like. Please give us a reasonable window to fix the issue before disclosing it publicly, and do not access other people's data while investigating.

What we do not have

The honest half of a security page.

Security pages usually list only the wins. Here is what we have not done yet, so it does not come as a surprise in a review.

  • No SOC 2 or ISO 27001 certification. We are an early company and have not been through either audit. We will not imply otherwise.
  • No third-party penetration test yet. We run our own automated attack suite against the agent guardrails on every change, but no external firm has assessed us.
  • No single sign-on or directory provisioning. Authentication today is email and password or Google. SAML and SCIM are not built.
  • Not assessed for regulated health data. Do not put material subject to HIPAA or comparable regimes into Onvoi.
  • Data is processed in the United States. There is no regional residency option yet.

If any of these blocks an evaluation, tell us — knowing which one stops teams is how it gets prioritised.

Bringing this to a security review?

We will walk your team through the data flows and answer the awkward questions directly.