Question 1

You call yourself security-first. What does that mean at the architecture level, not the marketing level?

Accepted Answer

Three things you can verify rather than take on faith. Connector credentials are stored as ciphertext under envelope encryption, with data keys wrapped by a per-tenant AWS KMS key that our application servers can decrypt but not export. Every workflow run writes an append-only audit record carrying actor, resource, and outcome. And each AI step executes inside a declared scope, so a summarisation step cannot call a payment API even if the model asks it to. What this is not: a certification. It is architecture and operating practice, and we publish the details so your team can judge them.

Question 2

Can you actually delete my data from your logs, or is that just a support ticket that goes nowhere?

Accepted Answer

Partly, and the honest answer matters more than the reassuring one. Deleting a record purges the payload from primary storage and from the searchable execution history, and the deletion itself is recorded. But audit entries stay append-only by design: we keep the fact that a run happened, its actor, resource, and outcome, because a log you can quietly rewrite is not an audit log. Encrypted backups roll off on a 35-day cycle, so a deleted payload can persist in backup for up to 35 days. If you need shorter than that, say so before you sign, not after.

Question 3

If AWS KMS goes down in your region, do my workflows just stop?

Accepted Answer

Workflows that need a connector credential stop, and that is deliberate. Data keys are cached in memory for a short window, so brief KMS latency is invisible; a sustained regional KMS failure means we cannot unwrap credentials, and we fail closed rather than fall back to a weaker key path. Affected runs enter a retry queue, hold their input payloads, and resume when key access returns. You get a status-page entry and a webhook alert, not silent data loss. The tradeoff is availability: we would rather your invoice workflow run an hour late than execute with a key we could not verify.

Question 4

What happens if your AI approves something it shouldn't?

Accepted Answer

It will, eventually, and the design assumes that. AI steps produce a proposed action, not a committed one. Every action carries a risk score, and anything above your configured threshold, or touching a category you have marked sensitive, routes to a named human approver before execution. Approvals expire; unapproved actions never run. If a low-risk action was wrong anyway, the run record holds the exact model input, the output, and the rule that let it through, so you tighten one threshold instead of guessing. Reversibility depends on the target system: a Slack message can be deleted, a captured payment cannot.

Question 5

Does the model see my customer data, and are my prompts used for training?

Accepted Answer

Yes, the model sees whatever fields your workflow passes it, which is why field selection is explicit rather than 'send the whole record'. You choose the fields; a redaction step can mask PAN, phone, and email patterns before the call. Prompts and completions are not used to train any model, ours or a provider's, and we contract for zero-retention inference where the provider offers it. We do keep the input and output in your run record, because that is what makes an AI decision auditable, and that record obeys your retention setting. A workflow that must never touch a model can run on rules only.

Question 6

Are you DPDP compliant?

Accepted Answer

No vendor can be, on your behalf. DPDP obligations sit with you as Data Fiduciary; we act as a processor, and the most we can do is make your obligations evidenceable. Concretely: consent state is stored per data principal and checked at run time, erasure and access requests get a tracked workflow with timestamps and an assigned approver, breach-notification drafts are pre-templated, and the audit trail exports as signed JSON your counsel can read. We are designed to support DPDP obligations, not to discharge them. Anyone telling you their platform makes you compliant is selling you a false sense of finished.

Question 7

Do you have SOC 2 or ISO 27001? Our procurement team will ask for the report.

Accepted Answer

We do not hold SOC 2, ISO 27001, HIPAA, or STQC attestation, and we will not imply otherwise on a sales call. What we can hand your security team is a completed CAIQ-style questionnaire, an architecture document covering key management and tenant isolation, our access-review and change-management procedures, a subprocessor list with regions, and read-only access to a sandbox tenant so your reviewer can test controls instead of reading claims. If a certificate is a hard procurement gate, tell us in week one; we would rather lose the deal there than at legal review after your team has spent six weeks on it.

Question 8

When my auditor asks for evidence, do I have to raise a ticket and wait on you?

Accepted Answer

No. Audit export is a self-serve function on every paid tier, not a favour. An admin filters by date range, actor, workflow, or outcome and exports to CSV or signed JSON, including denied and escalated attempts, which are usually what an auditor actually wants to see. Retention is 90 days on Team, 12 months on Business, and configurable up to seven years on Enterprise; past your retention window we cannot produce what we no longer store, so set it to match your own policy on day one. Exports are themselves logged, so you can show who pulled evidence and when.

Question 9

Are these real integrations, or thin webhook wrappers with a logo next to them?

Accepted Answer

Both kinds exist in this market, so the question is fair. Our connectors for Gmail, Slack, Razorpay, Shopify, Zoho CRM, Google Sheets and WhatsApp Business are typed: each declares its triggers, its actions, and the exact OAuth scopes it requests, and that scope list is visible before you authorise anything. We request read-only scopes wherever the action allows it. We are not a partner of, or endorsed by, any of these vendors; we call their public APIs like any other client. For anything off the list there is a generic HTTP and webhook step, which is a wrapper, and we label it as one.

Question 10

What happens when a token expires or a vendor changes their API mid-run?

Accepted Answer

Runs fail at the step, not silently at the end. A 401 marks the connection degraded, pauses dependent workflows rather than retrying into an account lockout, and notifies the connection owner by email and webhook. Transient 5xx and rate-limit responses retry with exponential backoff behind a per-connector idempotency key, so a retried payment action does not charge twice. Breaking upstream API changes are the honest weak point: we track deprecation notices and version-pin where the vendor supports it, but an unannounced change means we are fixing it after you notice. Our published target for a connector fix is two business days.

Question 11

Where does my data physically live, and does any of it leave India?

Accepted Answer

Primary storage, execution queues, and audit logs run in the AWS Asia Pacific (Mumbai) region, ap-south-1, with encrypted backups replicated to Hyderabad, ap-south-2. Both are Indian regions. Two honest exceptions: our error-monitoring and email-delivery subprocessors handle metadata outside India, and if you enable a model provider with no Indian endpoint, inference leaves the country. Both appear in the subprocessor register with region and purpose, and both can be switched off per workspace at the cost of that feature. Connector traffic goes wherever the third party's API terminates; we cannot change where Slack or Shopify answers a request.

Question 12

What does the free tier not include? I assume the useful parts are paywalled.

Accepted Answer

Largely correct, so here is the line. Free is 500 runs a month, one seat, three connectors, and seven days of audit retention. It does not include human-approval routing, SSO, custom risk thresholds, IP allowlisting, signed audit export, or any support SLA. Runs above the cap queue and then drop; we do not silently bill overage. It is a real environment for testing whether a workflow behaves, not a way to run production for free, and we would rather say that now than have you find out during an incident. Free workspaces are deleted 30 days after the last login.

Question 13

How exactly do you count a run? Could a retry loop hand me a surprise bill?

Accepted Answer

A run is one workflow execution from trigger to terminal state, regardless of how many steps it contains, so a twelve-step workflow costs the same as a two-step one. Retries inside a run are free; a run that a loop re-triggers is a new run, which is the case that can surprise you. The guards: a per-hour execution ceiling you set yourself, loop detection that halts any workflow re-triggering itself more than five times a minute, and an email at 80 percent of quota. Overage on Team is ₹0.40 per run, billed monthly and capped at your stated ceiling.

Question 14

What does enterprise support actually commit to in writing, beyond a promise to care?

Accepted Answer

In the contract, not the brochure: 24x7 coverage for Severity 1, a 30-minute first response, a named technical account manager with a scheduled monthly review, and a 99.9 percent monthly uptime target with service credits when we miss it. Support runs in English and Hindi through IST business hours; overnight Severity 1 is English-only, which is a real limit and we state it. Credits are credits, not consequential damages, and no SLA compensates you for a workflow you built wrong. Business tier gets a four-hour business-hours response without the account manager. Escalation paths and the on-call rota are documented at onboarding.

Question 15

We already run about 30 workflows on another automation platform. How bad is the migration, really?

Accepted Answer

Plan two to four weeks for twenty workflows, and expect one unavoidable manual step. Our importer reads the JSON export from most mainstream automation platforms and maps triggers, actions, and field mappings to our equivalents, flagging anything without a match rather than guessing at it. The part nobody can automate: OAuth refresh tokens are not exportable, so every connection has to be re-authorised by whoever holds that account. Budget an hour per connector owner. We also suggest running both platforms in parallel for a week with the old side's actions disabled, so you compare outputs before cutting over. Migration hours are included on Business and Enterprise.

The questions a security review actually asks

Security

AI approvals

Compliance

Integrations

Data residency

Pricing

Enterprise support

Migration

Still unanswered