Designed for trustworthy AI support.
Trust comes from showing your work. SupportGraph is built around source-backed answers, human approval, traceability, and responsible AI principles.
Our trust commitments
SupportGraph is built around source-backed answers, scoped data boundaries, and a clear audit direction. The capabilities below describe what is available today and what is on our near-term roadmap. Compliance certifications such as SOC 2 are part of our roadmap; we will publicly announce status as each milestone is completed.
Source traceability
Every AI draft cites the documents and passages that justify it. Citations are visible in-product and can be included in customer-facing replies.
Human-in-the-loop review
No AI reply is sent without a teammate explicitly approving it. The approval queue is the default state.
Data handling principles
Customer knowledge sources stay scoped to your workspace. We are designing for least-privilege ingestion and clear data boundaries.
Access controls
Role-based access for ingestion, draft review, and approvals is on the roadmap. Early access uses workspace-level access.
Audit trail direction
Designed for traceable history: who approved what, with which sources, and which edits shaped the final reply.
Reliability & evaluation
Confidence scoring, source coverage, and approval rate are built in so quality is observable, not assumed.
Responsible AI principles
Grounded answers. Citations. Approval. Bounded autonomy. We do not ship features that ask customers to trust the AI without evidence.
Customer data, not customer training
Designed so workspace knowledge is used to serve that workspace, not to train a shared model. We will document this clearly as we move to pilot.
Private-by-design architecture
Tenant boundaries, scoped credentials, and minimal data movement are core architectural goals from day one.
Principles we hold ourselves to
A short, honest set of rules that shape what we build and what we refuse to ship.
Ground first. If we cannot retrieve sources for a claim, we do not ship a feature that pretends we can.
Show the evidence. Citations are not a developer feature. They are how a support reviewer trusts the draft in front of them.
Bounded autonomy. Drafts go to a human. The human approves. We will earn the right to relax this only with measurable evidence.
Humble defaults. When confidence is low or sources are missing, the system says so instead of guessing.
Have a specific security question?
If your team is evaluating SupportGraph for production support, write to us. We'll share a security overview tailored to your stack.