Subprocessors
Last updated July 16, 2026
These are the service providers that process data on Semfora's behalf, what each one processes, and where their own legal terms live. How we handle your data overall — including source code — is described in our privacy policy. We give at least 30 days' notice before a new subprocessor begins processing customer data, by updating this page and announcing the change; email contact@semfora.ai to be notified of subprocessor changes directly, or to request our data processing agreement (DPA).
Cloudflare
Application hosting, edge caching, and object storage.
Data processed: Request logs (IP address, user agent), cached dashboard data, derived index artifacts (names and structure, never source text).
Trigger.dev
Isolated analysis compute — the ephemeral containers where your repository is cloned, analyzed, and deleted at the end of every run.
Data processed: Cloned source code (in memory and scratch disk for the duration of a run only), run logs without source content.
Prisma (Prisma Postgres)
Primary database.
Data processed: Account, organization, billing, and integration records; derived analysis metrics (names and numbers).
Clerk
Authentication, organizations, invitations, and billing.
Data processed: Names, email addresses, authentication credentials, subscription state.
Stripe (via Clerk Billing)
Payment processing.
Data processed: Card details and payment records (held by Stripe; Semfora never receives card numbers).
OpenRouter
AI inference for the features described in our privacy policy.
Data processed: Prompts built from analysis metadata; for explicitly-invoked code-reading features, relevant source excerpts (processed transiently — see the AI section below).
PostHog
Product analytics.
Data processed: Usage events tied to account identifier and email; never prompt, code, or AI response content.
Sentry
Error monitoring for Semfora's own application.
Data processed: Error reports with browser/request context (may include IP address); repository access tokens are scrubbed.
GitHub
Repository access via the GitHub App you install.
Data processed: Installation and repository metadata, webhook events, short-lived clone tokens.
Linear
Work-item sync, when you connect the Linear integration to your workspace.
Data processed: Work-item references (identifiers and titles) and the OAuth credentials you authorize; only used when you connect Linear.
Upstash
Rate limiting and replay protection.
Data processed: Short-lived request counters and nonces; no personal data or code.
AI models and enforced settings
All AI inference runs through OpenRouter. The models we currently use, and what for:
anthropic/claude-sonnet-4.5— Copilot chat (default), plain-English symbol labels, pull-request briefs, balanced code audits.anthropic/claude-opus-4.8— Health narratives and insights, AI dashboard generation, change briefs, deep code audits.anthropic/claude-haiku-4.5— Agent actions (default), error clustering, standard automated error diagnosis, fast code audits.
Zero Data Retention routing (zdr) is enforced on every request: OpenRouter only routes our traffic to provider endpoints that do not store prompts or responses at rest. See OpenRouter's ZDR documentation.
No-training routing (data_collection: deny) is enforced on every request: providers that use inputs to train models are excluded.
OpenRouter's own prompt logging is not enabled for our account; OpenRouter does not retain our prompts or completions.
These settings are set in code on every request our systems make, not just as account configuration, and this page is updated whenever they change.
Questions
Questions about a provider on this list? Email contact@semfora.ai.