Skip to main content
RoutMe

Your customers' conversations, handled like they matter

Messaging data is unusually sensitive. It contains order histories, phone numbers, health appointments, and whatever else a person decided to type into a chat. This page describes what we do with it, what the AI agent can reach, and what we have not done yet.

TLS 1.2+
Encrypted in transit

API calls, webhooks, dashboard sessions, and tool calls.

Never
Used to train a model

Conversations and connected data are processed to deliver the service, and nothing else.

2 days
To acknowledge a report

Business days, for any vulnerability sent to our security address.

Public
Data Processing Agreement

Published for anyone to read, not negotiated per deal.

Where your data goes

One path, four stops. The agent reads from your systems when it needs an answer, and nothing is copied out along the way.

  1. Your customer

    Messages on WhatsApp or Chat

  2. The channel API

    Arrives over the channel's official API

  3. RoutMe

    Flows, the agent, and the audit log

  4. Your systems

    Read at the moment of need, never copied

The baseline

The controls that are in place today, described plainly.

Encryption in transit and at rest
Everything between your systems and ours moves over TLS 1.2 or above: API calls, webhook deliveries, dashboard sessions, and agent tool calls alike. Stored message content and contact records are encrypted at rest by the managed data services we run on.
Scoped access, both ways
Team members get role-based access in the dashboard. API keys are scoped at creation as read or write, so a key that only reads cannot send.
Audit logging
API calls, configuration changes, flow edits, and administrative actions are logged with the actor and the timestamp. Conversation transcripts and flow run traces show which steps ran and which failed.
Monitoring and alerting
Infrastructure health, API error rates, and delivery patterns are monitored continuously. Anomalies in traffic shape are alerted on, because an unusual send pattern is often the first sign of a compromised key rather than a busy day.
Incident response
We have documented procedures for triage, containment, and notification. Affected customers are contacted directly, and anything with service impact is posted to the status page with a summary once it is resolved.
Data protection law
We apply GDPR obligations wherever they reach our customers, and operate under India's DPDP Act 2023 as an India-incorporated company. A Data Processing Agreement is published rather than negotiated per deal, so you can read the terms before you talk to anyone.

What the AI can and cannot reach

This is the question worth asking of any agent product, and most of the answer is architectural.

  • Your data is read, not absorbed

    The agent queries your systems at the moment it needs an answer and uses the result for that turn. We do not build a shadow copy of your database in order to answer questions about it.

  • It can only call what you attached

    Tools are declared per agent node. A node with two read tools has exactly two capabilities, and adding a third is a change you make deliberately rather than something the model can decide it needs.

  • Nothing you send trains a model

    Customer conversations and connected business data are not used to train models, ours or anyone else's. They are processed to deliver the service you asked for and for nothing else.

  • Credentials stay on your side

    Tool connections reference a secret you hold and call an endpoint you control. The narrower the endpoint, the less there is to lose, which is why read-only lookups are the default starting point.

  • Consequential steps sit behind your logic

    Refunds, cancellations, and writes back into your systems run as flow actions guarded by conditions you wrote. The agent proposes; the flow decides.

  • Tool output is treated as untrusted

    Anything returned from a connected system can contain text a customer typed. Return the fields you mean to return, and keep free-form blobs out of what the agent reads.

Where a message goes, from arrival to deletion

  1. A message arrives

    It lands over the channel's official API, is associated with a contact on your account, and is stored against that conversation. It is your data throughout: you are the controller, we are the processor.

  2. The agent resolves it

    Only the context needed for that turn is sent to the model, along with what your connected tools returned. Transcripts and flow run traces let you review the reply later.

  3. The reply is delivered

    Delivery state comes back from the channel and is recorded against the message. This is what your webhook receives and what the dashboard displays.

  4. Retention and deletion

    Conversation content is kept for the period described in our privacy notice so that logs, billing, and support remain usable. When it lapses, or when a deletion request comes in, the record is removed rather than archived.

Sub-processors

Delivering the service means other providers touch data: the messaging channel itself, cloud infrastructure, and the model provider behind the agent. The current list, what each one handles, and where it processes data are maintained in the Data Processing Agreement, which is public rather than gated.

We do not sell platform data, share it for advertising, or hand it to anyone outside that list.

Retention and deletion

Conversation content is retained for the window set on your account, then removed. Deletion requests are honoured for both our customers and the people they message, and the flow for making one is self-serve rather than an email thread.

Where a record must survive deletion for a legal or accounting reason, it is the billing metadata that persists, not the message body.

What we have not done yet

Plenty of security pages imply certifications that do not exist. Here is our actual position, so nobody discovers it during a procurement review.

Report a vulnerability

Send it to team@getvive.ai with enough detail to reproduce it. We acknowledge within two business days and keep you updated until it is closed.

  • Test only against accounts and numbers you control.
  • No denial of service, no social engineering, and no accessing another customer's data.
  • Give us a reasonable window to fix it before publishing.
  • We will not pursue legal action against research conducted in good faith under these terms.

Security questions

Something missing? Talk to us and a person will answer.

Evaluating RoutMe? We will answer the hard questions.

Send us your security questionnaire, or ask on a call. If something here is not good enough for you, we would rather know early.