Docs menu
Graphne
Ask about your graph in plain language, and control what the AI can see, do and spend.
7 guides
Use Graphne
- Ask GraphneAsk Graphne about your identity graph in plain language, get finding explanations, and read the Risk Score and blast radius scoring behind each answer.
- Audit Graphne’s workAudit every Graphne run: the model, system prompt, tool calls and graph queries it used, the data it saw, and whether people accepted or rejected its advice.
Set up and control AI
- AI providers and routingAdd LLM providers for Graphne, route tasks with failover and a circuit breaker, set model prices and a monthly budget, and send OpenTelemetry traces.
- AI safety controlsStop running AI work, turn AI off, restrict Graphne tools by role, check each agent's least-privilege access, and set a hard monthly spending ceiling.
- Decision modelsScreen attacker-written text for prompt injection with decision models, classifiers that give a probability for each answer before AI features read it.
- Decision assuranceCheck that a decision question is calibrated on your own data, attack-tested and drift-checked before it moves past shadow mode to advisory or automated.
- How Graphne’s queries stay read-onlyHow Graphne keeps the queries it writes read-only and bounded, defends against prompt injection, and what happens when a query is refused, truncated or slow.