sipi.bot vs OpenAI Moderation API: spend safety vs content safety
The OpenAI Moderation API flags harmful text. sipi.bot flags harmful transactions. If your agent can spend money, you need both — one protects your users, the other protects your wallet.
| Dimension | OpenAI Moderation API | sipi.bot |
|---|---|---|
| What it evaluates | Text — prompts and completions | Transactions — amount, merchant, category |
| What it blocks | Hate speech, violence, harassment, self-harm, sexual content | Over-budget spends, bad merchants, runaway loops, off-hours purchases |
| Decision model | ML classifier with category scores | Deterministic rules: caps, totals, velocity, allow/block lists |
| Output | Category flags with confidence scores (0-1) | APPROVED, BLOCKED, or FLAGGED with a human-readable reason |
| Configuration | Category thresholds per endpoint | Per-transaction limit, daily total, merchant allow/block, category limit, time window, velocity, approval threshold |
| Use case | Content safety — what the agent says | Spend safety — what the agent buys |
Two guardrails, not one
The Moderation API stops your agent from generating harmful text. It doesn't stop it from buying $2,000 of GPU time at 3am from an unknown vendor. sipi.bot stops the spend; the Moderation API stops the content. They're orthogonal safety layers and you need both when your agent talks to users and spends money.
The pattern: call the Moderation API on the agent's output to users, and call sipi.bot on the agent's spend actions. Two APIs, two threat surfaces, one safe agent.
Frequently asked
Does sipi.bot replace the OpenAI Moderation API?
No. The Moderation API is for content safety — catching harmful text. sipi.bot is for spend safety — catching harmful transactions. Use both when your agent interacts with users and spends money.
Can I call both in the same agent pipeline?
Yes. A typical flow: agent generates a response → Moderation API checks it for harmful content → agent proposes a spend → sipi.bot checks if it's allowed → agent executes the spend. Two guards, two checks, no overlap.
Is sipi.bot's rule engine as accurate as the Moderation API's ML?
Different accuracy models. sipi.bot's rules are deterministic — they fire exactly when the condition matches, with no false positives or false negatives. The Moderation API uses probabilistic ML. Both are appropriate for their domain.