Platform

Guardrails

Find email addresses, card numbers and credentials in prompts, and redact or block them before anything leaves LaunChad.

curlPUT /api/console/guardrails {"mode":"redact"}

Redact. Matches are replaced and the rest is sent.

Block. The request is refused with guardrail_blocked.

Everywhere. Chat, embeddings and agent model calls.

Redaction

Secrets stop
before they leave

Guardrails scan every message and text part. Card numbers are checksum-validated, so order numbers are left alone. What was redacted comes back in a response header.

  • Emails Addresses in any message.

  • Card numbers Only Luhn-valid numbers.

  • Credentials Private keys and common API key formats.

guardrails: redact
in:  my key is sk-proj-9xK2…out: my key is [redacted secret]x-launchad-guardrail: redacted; secret=1

Also included

The details
that matter

Off by default

Turn it on per workspace.

Audited

Mode changes are written to the audit log.

Linear-time

Patterns are built to resist ReDoS.

No prompt storage

Scanning happens in memory only.

Clear errors

400 guardrail_blocked names the rule.

Honest scope

Pattern checks, not a full DLP system.

Related

launchad · getting started

# 1. point your OpenAI client at LaunChad
export OPENAI_BASE_URL=https://api.launchad.si/v1
export OPENAI_API_KEY=$LAUNCHAD_KEY

# 2. call any model, or start an agent run
curl $OPENAI_BASE_URL/agents/researcher/runs \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{"message":"What changed on our competitors' pricing pages?"}'

Get started

Two lines of config.
The rest is built.

Keep your OpenAI client, change the base URL and the key. Limits, logs and guardrails are on from the first request.

Free during the beta. No credit card. Launch updates only, no spam.