sdlc.cc

Pre-release. sdlc.cc is not generally available. There is no completed customer deployment and no design partner yet, and per-class detection accuracy has not been published. See the evidence boundary.

JOURNEY_SENTINEL_HOME · PRE-SEND AI PRIVACY

Stop sensitive data at the moment of send.

sdlc.cc applies your organization’s policy before a prompt reaches ChatGPT, Claude, or an LLM API. Browser Guard now scrubs and continues the protected send in one action.

Build your protected package Watch the product moment

Supported today for managed Chrome and Edge on ChatGPT and Claude text sends. Read the support matrix.

One protected send. A synthetic prompt is checked, transformed, and sent without a second click. The clip uses illustrative data and does not claim measured detection accuracy.

01

One boundary. Five ways in.

Start where sensitive content is created. Keep one policy, one detector chain, and one evidence model as protection expands.

  1. A

    Browser Guard

    Protect supported ChatGPT and Claude text sends in managed Chrome and Edge.

    Set up Browser Guard
    Supported MVP
  2. B

    Gateway API + SDK

    Inspect or transform application traffic before the provider call.

    Set up the gateway
    Supported
  3. C

    Private Appliance

    Run the same engine and policy model inside customer-controlled infrastructure.

    Set up privately
    Supported
  4. D

    Claude Code gateway

    Install the protected Anthropic base-URL path on Windows, macOS, or Linux.

    Explore Claude Code protection
    Supported
  5. E

    Codex gateway

    Route supported OpenAI-shaped text calls through a scrub-forward preview boundary.

    Explore the Codex Preview
    Preview

02

Every supported model write crosses one governed path.

The system decides before dispatch. Adjacent and malformed routes are refused rather than passed through unprotected.

  1. 01AuthenticateResolve the credential and tenant.
  2. 02Apply policySelect the effective organization and group rule.
  3. 03TransformAllow, warn, mask, redact, tokenize, or block.
  4. 04Dispatch onceOne provider attempt after durable admission.
  5. 05Record safe evidenceCounts and decision metadata, never raw prompt text.

03

The audit schema has nowhere to put your prompt.

That privacy claim is structural. Evidence tables hold tenant scope, policy references, decision codes, detector counts, and timing. They do not define a column for prompt or completion text.

Inspect the architecture and evidence contract
privacy.decision.v1content omitted by design
tenant_id
tnt_acme
policy_version
pol_0042
decision
redact
detector_counts
{ email: 1, pan: 1 }
provider_attempts
1
prompt_text
not in schema

04

Investment-ready means the gaps are visible too.

Repository evidence is strong. Customer evidence is the next milestone, and the product does not borrow credibility from claims it has not earned.

Demonstrable now

  • Pre-send Browser Guard flow for supported ChatGPT and Claude text sends
  • Fail-closed supported model-write contract
  • Metadata-only evidence and tenant-scoped policy
  • Gateway API, installable JavaScript client, and private appliance

Still to earn

  • Named design-partner outcomes and willingness to pay
  • Published per-class detection results from a real gateway run
  • Store-approved distribution and production availability evidence
  • Self-serve entitlement after payment

Read the complete claim-truth boundary

05

Give one team a safer way to use AI.

Begin with a bounded protected-team pilot, tune the policy with real workflows, and keep the evidence honest from day one.

Start the pilot conversation See pilot and team pricing