Platform · InspectionEvery prompt, response, and tool-call — inspected in real time

18ms. Every call.
No exceptions,
no blind spots.

Obiguard's inspection pipeline sits in the critical path between your application and the model. Every token in and every token out passes through 64 detectors before the call completes.

01 / How inspection works

Synchronous. In-path. Stateful.

Unlike async log-based tools, Obiguard evaluates policy before the response is returned. A blocked call never reaches the model — or the user.
01 — PROMPT

Prompt inspection

Every user prompt and system message is scanned for injection attempts, policy violations, PII, and off-scope content before the request reaches the model.

64 detectors · prompt & system
02 — RESPONSE

Response inspection

Model completions are inspected before delivery. Obiguard can redact, block, or flag content — with the original preserved in the audit ledger.

Redact · Block · Flag
03 — TOOL CALLS

Tool-call validation

Agent tool invocations are validated against your registered allow-list. Calls to out-of-scope tools are blocked — logged with full context.

~5ms added latency
02 / Detectors

64 detectors. Continuously updated.

Detectors cover PII (12 types), prompt injection, toxicity, off-topic drift, model jailbreaks, data exfiltration patterns, and custom regex or semantic rules you define.
64
Detectors shipped

Covering PII, injection, toxicity, exfiltration, and model-specific jailbreaks.

18ms
p50 added latency

Synchronous, in-path inspection — not a background log processor.

12
PII types

SSN, credit card, passport, NHS, IBAN, email, phone, and more — all redactable.

7
Frameworks mapped

Every detector maps to at least one compliance framework control.