Introduction

curl -X POST https://api.getsieve.dev/v1/moderate/text \
  -H "Authorization: Bearer mod_live_xxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "you absolute noob get rekt lmao",
    "context": "gaming_chat",
    "user_id": "player_42"
  }'

{
  "id": "mod_r_8f3k2j",
  "flagged": false,
  "action": "allow",
  "tier_used": 0,
  "latency_ms": 3,
  "categories": {
    "toxicity": { "score": 0.32, "threshold": 0.7, "flagged": false },
    "harassment": { "score": 0.18, "threshold": 0.6, "flagged": false },
    "profanity": { "score": 0.41, "threshold": 0.8, "flagged": false }
  }
}

That request took 3ms and cost $0. The local Rust pipeline handled it entirely on the edge — no AI API call needed. “Get rekt noob” is standard gaming banter, and Sieve’s gaming mode knows the difference.

---

What is Sieve?

Sieve is a content moderation API built for developers. One endpoint, one API key, and your app has production-grade moderation that understands context.

We handle the hard parts — evasion detection, context-aware scoring, custom rules — so you can focus on building your product.

Local Rust processing handles 60-70% of requests at zero cost. AI tiers activate only when needed.

Tier 0 runs Unicode normalization, leetspeak decoding, and pattern matching in Rust/WASM on the edge. Most requests never leave.

Built-in context for gaming chat, usernames, and in-game communication. Knows that “gg ez” isn’t harassment.

Migrating from Google Perspective API? Swap one URL. Same request format, same response shape, better results.

The Problem

Content moderation is broken for most developers:

• Too expensive. AI moderation APIs charge per-request, and costs explode with scale.
• Too slow. Round-trips to AI models add 200-500ms of latency to every message.
• Too dumb. Generic models flag gaming banter as toxic and miss actual abuse wrapped in Unicode tricks.
• Too complex. Building your own pipeline means maintaining regex lists, training models, and handling evasion — forever.

Sieve fixes this with a tiered pipeline that’s fast by default, smart when it needs to be, and priced so you don’t think about it.

Who is Sieve For?

In-game chat, usernames, clan names, profile bios. Sieve understands gaming context and won’t over-flag your players.

Posts, comments, threads, user profiles. Configurable thresholds per content type with custom category support.

Discord bots, Twitch bots, Slack apps. Low-latency moderation that keeps up with real-time chat.

Key Concepts

Concept	What it means
Tiers	Three processing levels (local, standard AI, advanced AI) — requests escalate only when needed
Context	Tell Sieve what kind of content it’s looking at (gaming_chat, forum_post, username, comment) to get context-aware scoring
Categories	Each request returns scores for categories like toxicity, harassment, profanity, sexual, self_harm
Actions	Based on scores and your thresholds: allow, flag (for review), or block
Shadow mode	Run Sieve alongside your existing moderation to compare results before switching over

Next Steps

Make your first API call in under 2 minutes.

Understand the 3-tier pipeline architecture.