§ Security

Security you can verify.

Discera processes sensitive sales-call data. We take that seriously — strong encryption, strict access controls, and a formal compliance program that we make legible instead of hiding behind a one-pager.

discera · security statusas of 2026
  • SOC 2 Type IAssessment underway
  • SOC 2 Type IIOn roadmap
  • Encryption at restAES-256 · Fernet
  • Encryption in transitTLS 1.2+ · HSTS
  • Data residencyGCP us-central1
  • Vuln disclosuresecurity@discera.ai
Live·In progress·Roadmap
§ Controls

Six layers of protection

Security isn’t one decision — it’s a stack. Here is every layer we have in place today.

Compliance
SOC 2
  • Type I assessment underway (Security, Availability, Confidentiality)
  • Type II roadmap in progress
  • Control matrix and evidence documented in-house
  • Enterprise DPA, MSA, and security review available on request
Encryption
Data at rest & in transit
  • Cloud SQL and GCS encrypted with Google-managed AES-256 keys
  • Gong and HubSpot credentials encrypted with Fernet before storage
  • Secrets stored in Google Secret Manager — never in source code or env files
  • TLS 1.2+ enforced on all connections; HSTS on all public endpoints
Infrastructure
Google Cloud Platform
  • Data resides exclusively in GCP us-central1 — no cross-region data movement
  • All services run on Cloud Run inside a private VPC (discera-vpc)
  • No persistent VM access; immutable container deploys only
  • Cloud SQL with PITR enabled; private IP, no public exposure
Access control
Authentication & roles
  • Clerk-managed authentication with short-lived session tokens
  • Organisation-scoped data — workspaces are fully isolated at the database level
  • Two roles: admin (billing, integrations, settings) and member (analysis, reports)
  • Two-factor authentication available for all users; enforcement configurable per workspace by admins
Data handling
Isolation & retention
  • Each workspace is a Clerk Organisation — workspace isolation enforced at the database query level
  • Gong and HubSpot OAuth tokens encrypted at rest; never logged or exposed in responses
  • Call transcripts fetched on-demand for analysis and processed in memory; not persisted to our database
  • Data deletion on workspace closure; enterprise retention schedules available
Vendor security
Subprocessors
  • Full subprocessor list at /legal/subprocessors
  • Google Cloud Platform — SOC 2 Type II, ISO 27001
  • Anthropic, Clerk, Stripe, OpenAI — each SOC 2 certified
  • Vendor security reviewed before onboarding; inventory and DPA status tracked annually
§ AI processing

Your calls don’t train anyone’s models.

Discera routes transcript data through Anthropic and OpenAI for per-call analysis and executive summaries. This is the number-one question we expect from security reviewers, so we’re stating it plainly: your call data is never used to train or improve any model.

No model training on your data

Both Anthropic and OpenAI explicitly prohibit use of API inputs/outputs for model training under their commercial terms. This is a contractual commitment, not a default setting — our data processing agreements with each provider govern it.

Anthropic (claude-sonnet-4-6)

Per-call analysis runs on Claude. Anthropic's commercial terms (effective 2026-01-01, including SCCs) explicitly exclude API data from model training. DPA executed and on file.

OpenAI (GPT-4o / GPT-4o-mini)

Executive summaries run on OpenAI. OpenAI's API usage policies exclude API data from model training by default. Enterprise customers can additionally enable zero data retention — available on request.

Transcripts processed in memory

Transcript content is fetched from Gong on-demand, sent to the LLM provider for the analysis request, and not persisted to Discera's database. What we store is the structured analysis output, not the raw transcript.

§ Infrastructure

No persistent servers.

Every service is an immutable container on Cloud Run. There are no long-lived VMs, no SSH keys floating around, and no manual production access paths.

Cloud Run (container-native)

All services — API, worker, MCP server, chat — deploy as immutable container images. New code ships as a new revision; rollback is instant. No SSH, no persistent compute.

Private VPC network

Backend services communicate over discera-vpc. Cloud SQL (Postgres 15) and Memorystore (Redis) are private-IP only — unreachable from the public internet.

Secret Manager

Every secret — API keys, signing keys, encryption keys — is stored in Google Secret Manager and injected at runtime. Secrets are never committed to source control or baked into images.

Cloud SQL with PITR

Postgres 15 with point-in-time recovery enabled. PITR restore procedure is documented in a runbook and exercised on major releases.

§ Access control

Workspace-scoped, always.

Every report, credential, and job is tagged to a Clerk Organisation. The database enforces workspace isolation at the query level on every API call.

Clerk authentication

Short-lived JWTs issued by Clerk on every request. Tokens are verified server-side on every API call. The backend never stores session tokens.

Two roles: admin and member

Admins manage billing, integrations, and workspace settings. Members run analyses and view reports. Audit logs (Scale+) record every sensitive action by actor, timestamp, and IP.

Two-factor authentication

Two-factor authentication is available for all users via Clerk. Workspace administrators can enforce 2FA org-wide from the settings panel. SAML SSO is available for Enterprise workspaces on request.

OAuth tokens encrypted at rest

Gong and HubSpot OAuth access and refresh tokens are encrypted with Fernet before being written to the database. The encryption key is stored in Secret Manager, not alongside the data.

§ Responsible disclosure

Found something? Tell us.

We welcome good-faith security research on Discera’s production systems. If you discover a vulnerability, please report it to us privately before public disclosure.

security@discera.ai

PGP key & machine-readable policy at /.well-known/security.txt

  • 01
    Email us first

    Send details to security@discera.ai. Include steps to reproduce, impact, and any proof-of-concept. Encrypt with our PGP key if the details are sensitive — key available on request.

  • 02
    48-hour initial response

    You will receive a response within 48 hours acknowledging receipt. We will keep you updated on our investigation and resolution timeline.

  • 03
    Safe harbour

    We will not take legal action against researchers who follow responsible disclosure. This covers testing on your own account, minimal data access needed to confirm the issue, and stopping on confirmation.

  • 04
    Out of scope

    Social engineering, physical attacks, denial-of-service, and automated scans that generate excessive load are out of scope and are not covered by this policy.

§ Vendors

Subprocessors

Discera uses a small set of subprocessors, each selected for their own security posture. All are SOC 2 certified. The full list is kept up to date at /legal/subprocessors.

  • Google Cloud Platform — infrastructure
  • Anthropic — LLM analysis
  • OpenAI — LLM analysis (fallback / exec summary)
  • Clerk — authentication
  • Stripe — billing

SOC 2 Type I report available under NDA · request via email

§ Questions

Get in touch

Security questions, enterprise security reviews, DPA requests, or anything else compliance-related — reach us at:

Response within 1 business day for general enquiries · 48h for security reports