Security at Afora

Effective date: July 10, 2026 · Last updated: August 29, 2026 · Version 2.0

Afora gives you an AI agent with real access — to a machine of its own, and to the accounts you choose to connect. That makes isolation and credential handling the heart of our security work. This page describes, plainly, how we protect your data. Questions or concerns: security@aforademo.com.

Tenant isolation — one customer, one instance

• Every customer's agent runs on its own dedicated instance: a separate environment with its own operating-system identity, its own data directory, and its own agent and console processes.
• One customer's agent cannot read another customer's files, conversations, or credentials — isolation is enforced by the operating system, not just by application code.
• Your workspace, memory, and chat history live inside your instance, nowhere else.

Encryption and credentials

• All traffic is encrypted in transit with TLS 1.2+; HTTPS is enforced everywhere.
• Sign-in to Afora is handled by Clerk; payment details are handled by Stripe and never touch our systems.
• You never give Afora a password for your AI provider or connected accounts: connections use the provider's own OAuth sign-in, and we hold only the resulting token.
• Where the web app stores a credential (for example, an API key you provide), it is encrypted at the application layer before storage. Tokens held on your instance live under that instance's own isolation.
• Disconnecting a provider or account in the console retires its token.

Infrastructure and subprocessors

The web app runs on Vercel; agent instances run on Hetzner cloud servers in the EU (Germany); the application database is Supabase; authentication is Clerk; billing is Stripe. Each vendor is listed with its purpose on our Subprocessor list, maintains its own independent security program, and is bound by contract to use customer data only to provide its service to Afora.

AI data handling

• Your agent thinks with your own Claude or ChatGPT subscription. Content your agent sends to that provider is governed by your agreement and settings with them — check your provider's data controls. Afora adds no secondary use of its own.
• Afora does not train models on your data, and does not permit its own subprocessors to.
• AI processing for your instance uses only your instance's data.

Agent action safety

An agent that can act is an agent that can be attacked through what it reads. We treat that as a first-class design constraint:

• Your agent can act only through connections you granted, and every connection can be revoked at any time — at the provider or in the console.
• Content the agent encounters (web pages, inbound messages, documents) is treated as data to work on, not as instructions with your authority; manipulation attempts (prompt injection) are an explicit part of our threat model and our testing.
• The console shows what your agent is doing and has done, so its actions are inspectable rather than invisible.
• Instances are isolated from each other, so a compromised input in one workspace has no path into anyone else's.

Compliance posture

We are a small team and we describe our posture honestly:

• SOC 2 Type II: not yet certified. Our internal controls (written security policies, access control, incident response, vendor management, retention) are documented and available to customers under NDA on request.
• We complete customer security questionnaires (CAIQ/SIG-Lite style) for prospective customers.
• Afora personnel do not read customer workspace content except with your permission, for security investigation, or as required by law — and such access is logged.

Data durability and retention

• The application database is backed up automatically with point-in-time recovery; instance data is backed up on a fixed schedule. Backups age out on schedule.
• On cancellation you can export your data for 30 days; production data is deleted within 60 days after that (details in the Privacy Policy §8).

Incident response

We maintain a written incident-response plan. If an incident affects your data, we will notify you without undue delay — within 48 hours of confirmation for personal-data breaches — with what we know, what we've done, and a contact who can answer questions.

Reporting a vulnerability

We welcome good-faith security research. Report issues to security@aforademo.com (or see /.well-known/security.txt). Include steps to reproduce; we will acknowledge within 3 business days, keep you informed, and not pursue legal action for good-faith research that respects user privacy and avoids service disruption. Please do not access data that isn't yours or degrade the service. No bug bounty is offered at this time; we credit reporters who want it.