Reliability and security

An enquiry lost on the way is a customer lost. Leadveyor is built so that every message is stored before anything else happens, every step that can fail is retried, and anything that still fails stays visible and can be run again.

Store first, then process

When a message arrives — by email, form or API — it is saved and acknowledged first, then placed in a processing queue. The sender doesn't wait for the AI steps, and a problem in a later step can never lose the message itself. A message identity (email Message-ID, form idempotency key, your externalId) makes sure a message received twice is processed once.

Retries

Processing a message
5 attempts · ~2.5 min
0
10s
30s
70s
150s
Delivering a webhook
8 attempts · ~2 h
0
1m
3m
7m
15m
31m
63m
127m
Attempts with growing pauses. A message or delivery that still fails is kept, marked failed, and can be run again.

Processing a message is tried up to five times, with pauses of 10, 20, 40 and 80 seconds — about two and a half minutes — which rides out short outages of an AI or maps service. Delivering a webhook is tried up to eight times, with pauses from 1 to 64 minutes — about two hours.

  • Only temporary failures (a service unavailable, a timeout) are retried; a message that can't be processed for a lasting reason fails at once with an explanation.
  • A message that fails after all attempts is kept with status failed and the reason, raises an inbound.failed event, and can be processed again from the Inbox or the API.
  • Each processing run is recorded with its steps, timings, models used and cost.

Webhook delivery

  • Signed — every request carries a Leadveyor-Signature header, an HMAC-SHA256 of the timestamp and the body with your secret, so your server can reject anything that isn't from Leadveyor and replays of old requests.
  • At least once — a delivery may be repeated after a timeout; each event has an ID (Leadveyor-Event-Id) to ignore duplicates.
  • Logged — every attempt with your server's response is kept and visible; any delivery can be sent again.
  • Self-protecting — an endpoint that fails ten deliveries in a row is disabled, with the reason shown, until you enable it.

Access

  • People sign in to the console; their role in the account (owner, admin, member) decides what they can change.
  • Programs use API keys with a permission — read, write or admin — limited to one workspace or the whole account. Keys are shown once and stored only as a hash; they can be revoked instantly.
  • The same rules everywhere — the console, the API and integrations go through one set of permission checks.

Separation

  • Workspaces never share data: leads, contacts and matching stay within a workspace.
  • Test and live data are kept apart: test keys create test leads, which never merge with live ones and only go to test webhooks.

Messages and models

  • Messages are read by language models only to classify, extract and answer your criteria; they are passed as data, never as instructions, and quotes are verified against the original text.
  • Incoming email is accepted only with a valid signature from the mail provider.
  • Passwords are stored as salted hashes; sessions and keys can be revoked.

Was this page helpful?