Comparison
Bird vs Prelude.
Prelude is a verify-only API with one of the sharpest fraud-detection models in the category. We're a broader platform with verify as one resource among nine. The honest read is below — with the matrix, the code, and where each one wins.
Switching Costs
Architecture
Compare with
What Prelude is great at.
Where Bird is different.
What Prelude is great at
Hyper-specialized verify routing.When the entire company is one product, the routing logic, the carrier deals, and the failover paths all get the attention a general-purpose platform can't give one resource.
ML-based fraud detection at create time. The model scores the verify attempt before it sends. Bot-driven verify abuse, SIM-farm signals, geo anomalies — blocked at create, not refunded after. The depth on this signal is the headline feature.
Real-time carrier-aware delivery optimization for OTP. Routes are picked per carrier, per country, with feedback loops tuned for OTP latency and conversion, not for cost across general A2P SMS.
Transparent per-verify pricing. One unit, one number, in their docs. Easy to model.
Single-product focus. The whole team works on one thing. The pace of iteration on verify specifically is faster than a generalist platform can manage by definition.
Where Bird is different
Verify is one resource in a broader platform. bird.verifications.start lives next to bird.emails.send, bird.sms.send, bird.voice.calls.create, bird.whatsapp.send. Same auth, same webhook envelope, same error shape.
One vendor, fewer contracts.If you also send transactional email, account notifications by SMS, or marketing pushes — you don't add another vendor to procurement, another set of credentials to rotate, another webhook handler to maintain.
Fallback chain across channels. bird.verifications.start({ channel: "sms", fallback: "voice" })is one call. If you wanted Prelude verify to fall back to a voice-call OTP from a different provider, you'd be writing that orchestration yourself.
MCP-first for agents. Official MCP server with verify tools surfaced. Agent-friendly without you writing the wrapper.
The matrix
Nine capabilities, side by side.
Prelude wins on verify-specific depth; Bird wins on scope and consistency across resources. Both are accurate today.
| Capability | Bird | Prelude | Who wins? |
|---|---|---|---|
| Verify-only product depth | Solid. SMS, voice, email channels. Per-recipient rate limits. Fraud score on the start response. Fallback chain in one attribute. | Excellent. The entire product is verify. ML fraud detection at create time, carrier-aware delivery optimization just for OTP, dispute analytics, hyper-tuned routing logic per region. | |
| ML fraud detection at create time | Risk score on the start response. Per-recipient rate limits enforced before send. Less granular than a specialist. | Real-time scoring model trained specifically on verify fraud signals. Blocks at create time, not after. The depth on this one signal is the best we've seen. | |
| Carrier-aware OTP routing | Operator-grade routing across 240 direct-to-carrier SMS connections. Tuned for delivery and cost across all SMS, not exclusively OTP. | Routing logic optimized specifically for OTP latency and conversion. The whole stack is tuned for one use case. | |
| Pricing transparency | From $0.04 per verification, all-in. Volume discounts auto-apply. | Per-verify pricing, published. Transparent enough to model in a spreadsheet. | |
| Other channels on the same API | bird.verifications.start lives next to bird.emails.send, bird.sms.send, bird.voice.calls.create, bird.whatsapp.send — one auth model, one webhook envelope. | Verify only. Sending a transactional email or a marketing SMS is a different vendor. | |
| Webhooks and error envelope | Same HMAC-SHA256 signature header and same error envelope for verify as for every other Bird resource. | Specific to the verify product. Clean within its scope; doesn't cross over to other channels because there are none. | |
| MCP / agent surface | Official MCP server. Tools include start_verification, check_verification, get_message_status. | No public MCP server today. OpenAPI spec available for agents that can read it. | |
| Single-product focus | Verifications is one of nine resources. Care about it; not the only thing we ship. | Verify is the only thing they ship. That focus translates into faster iteration on this one product than any generalist platform can manage. | |
| Compliance and rate-limit defaults | Per-recipient rate limits, lockout after configurable attempts, expiry default 10m. Reasonable defaults; auditable from the API. | Defaults tuned for verify abuse patterns specifically. Stricter out-of-the-box; more knobs for verify-specific tuning. |
The same verify start
Same conceptual model, slightly different field names.
Creating a verification with fraud-signal context. Prelude uses verification.create with a target object; Bird uses verifications.start with to and channel. Both accept a signals object for device, IP, and user-agent context. Eighteen lines vs sixteen — effectively the same shape.
Prelude
import Prelude from "@prelude.so/sdk";
const prelude = new Prelude({
apiToken: process.env.PRELUDE_API_TOKEN,
});
const verification = await prelude.verification.create({
target: {
type: "phone_number",
value: "+15005550006",
},
signals: {
device_id: "device_abc",
ip: req.ip,
user_agent: req.headers["user-agent"],
},
});
console.log(verification.status);
// → "success" | "blocked" | "retry"
// 18 lines.Bird Verifications
import { BirdClient } from "@bird/sdk";
const bird = new BirdClient({ apiKey: process.env.BIRD_API_KEY! });
const { data, error } = await bird.verifications.start({
to: "+15005550006",
channel: "sms",
fallback: "voice",
signals: {
device_id: "device_abc",
ip: req.ip,
user_agent: req.headers["user-agent"],
},
});
if (error) throw error;
console.log(data.id, data.risk_score);
// → "ver_2bX91Yk8h" 0.04
// 16 lines. Slightly different field names; same conceptual model.Switching cost
Low.
Same conceptual model on both sides: create a verification, hand the user the OTP, check the code. Field names differ slightly — target.value vs to; signals on both, with overlapping shape. The webhook envelope differs in signature header format only.
Realistic timeline: one to three engineering days.Dual-run for a week with both keys live, watch the success-rate graphs, cut over. If you'd built custom logic against Prelude's fraud-block reasons, plan for an afternoon to map those onto Bird's risk_score threshold.