ServerMessageEndOfCallReport can't parse payloads from Voice Agent because camelCase fields aren't honored

[pratyushsahay]

Description
When receiving a payload from a Vapi Voice Agent, the ServerMessageEndOfCallReport model in the server-sdk-python fails to parse the incoming JSON. The payload uses camelCase field names (e.g., endedReason, endTime, secondsFromStart), but the SDK model expects snake_case (ended_reason, end_time, seconds_from_start).

Although the generated model defines aliases using FieldMetadata(alias="..."), those aliases are not actually used during model validation. As a result, the SDK throws large validation errors stating that required fields are missing.

---

Example

from vapi_server_sdk.models import ServerMessageEndOfCallReport

msg = {
    "timestamp": 1762133576209,
    "type": "end-of-call-report",
    "endedReason": "customer-ended-call",
    "analysis": {"summary": "...", "successEvaluation": "true"},
    "artifact": {
        "messages": [
            {
                "role": "bot",
                "message": "Hi, this is a test.",
                "time": 1762133501000.0,
                "endTime": 1762133502000.0,
                "secondsFromStart": 1.0,
            }
        ],
        "transcript": "AI: Hi, this is a test."
    }
}

ServerMessageEndOfCallReport.model_validate(msg)

Actual behavior

80 validation errors for ServerMessageEndOfCallReport  
ended_reason  
  Field required [type=missing, input_value={...}, input_type=dict]  
artifact.messages.0.BotMessage.end_time  
  Field required [type=missing, input_value={...}, input_type=dict]  
artifact.messages.0.BotMessage.seconds_from_start  
  Field required [type=missing, input_value={...}, input_type=dict]  
...

---

Expected behavior
The SDK should correctly parse the Voice Agent’s camelCase JSON payload, populating the corresponding snake_case model fields using Pydantic’s alias handling.

---

Root cause
The generated models use Fern’s custom FieldMetadata(alias="...") annotations instead of Pydantic’s native Field(alias="...").
Pydantic v2 does not honor those aliases when parsing input (model_validate, __init__), only when serializing output (model_dump(by_alias=True)).

This means that incoming camelCase payloads from the Voice Agent (which follow the documented API schema) cannot be deserialized without manual key remapping.

---

Workarounds

Convert the incoming JSON keys from camelCase → snake_case before validation:

import re  
from typing import Any, Mapping

CAMEL_RE = re.compile(r'(?<!^)(?=[A-Z])')

def camel_to_snake(name: str) -> str:
    return CAMEL_RE.sub('_', name).lower()

def normalize_keys(obj: Any) -> Any:
    if isinstance(obj, Mapping):
        return {camel_to_snake(k): normalize_keys(v) for k, v in obj.items()}
    elif isinstance(obj, list):
        return [normalize_keys(x) for x in obj]
    else:
        return obj

normalized = normalize_keys(msg)  
ServerMessageEndOfCallReport.model_validate(normalized)

Or manually subclass the model to redeclare fields using pydantic.Field(alias="...").

---

Proposed fix

• Update code generation for server-sdk-python to use pydantic.Field(alias=...) rather than FieldMetadata(alias=...), ensuring proper alias resolution during validation.
• Alternatively, update the SDK’s UncheckedBaseModel to include:

model_config = ConfigDict(populate_by_name=True)

and propagate real Pydantic alias definitions into the generated models.

---

Environment

• SDK: vapi/server-sdk-python
• SDK Version: 1.7.3
• Pydantic: 2.12.3
• Python: 3.11
• Platform: macOS

---

Impact
Payloads received from Voice Agents fail validation because camelCase fields aren’t recognized. This breaks out-of-the-box use of the server SDK for handling voice agent events.