"""
Copyright 2026 OÜ KAVAL AI (registry code 17393877)
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
"""
import os
import time
import json
from typing import Any, Dict, List, Optional, Tuple, Type
from anthropic import AsyncAnthropic
from pydantic import BaseModel
from kavalai.llm_clients.base_client import (
BaseLlmClient,
ChatHistory,
LlmClientException,
LlmClientParameters,
ModelCallStat,
ModelStatsReceiver,
)
from kavalai.llm_clients.streamer import Streamer
# The Messages API requires max_tokens on every request. It is only an output
# ceiling (nothing is pre-allocated) and responses stream, so a generous
# default is safe.
DEFAULT_MAX_TOKENS = 64000
[docs]
class AnthropicClient(BaseLlmClient):
"""
Anthropic (Claude) LLM client implementation using the Messages API and
Streamer.
"""
def __init__(
self,
model: str,
llm_client_parameters: Optional[LlmClientParameters] = None,
model_stats_receiver: Optional[ModelStatsReceiver] = None,
api_key: Optional[str] = None,
base_url: Optional[str] = None,
):
"""
Initialize the Anthropic client.
Args:
model: The Anthropic model name (e.g., 'claude-opus-5').
llm_client_parameters: Optional parameters like temperature, top_p, etc.
model_stats_receiver: Optional receiver for model call statistics.
api_key: Optional API key (falls back to ANTHROPIC_API_KEY env var).
base_url: Optional base URL for the API.
"""
super().__init__(llm_client_parameters, model_stats_receiver)
self.model = model
self.api_key = api_key or os.getenv("ANTHROPIC_API_KEY")
self.base_url = base_url
timeout = 30.0
if self.parameters and self.parameters.timeout_seconds:
timeout = self.parameters.timeout_seconds
self.client = AsyncAnthropic(
api_key=self.api_key, base_url=self.base_url, timeout=timeout
)
async def _run_chat_completions(
self,
chat_history: ChatHistory,
response_model: Optional[Type[BaseModel]],
streamer: Streamer,
):
"""
Background task to handle the actual Anthropic API call and stream results.
"""
start_time = time.perf_counter()
value_streamer = streamer.get_value_streamer(
"response", response_model=response_model
)
system_prompt, messages = convert_messages(
[msg.model_dump() for msg in chat_history.messages]
)
call_kwargs: Dict[str, Any] = {
"model": self.model,
"max_tokens": DEFAULT_MAX_TOKENS,
"messages": messages,
}
if system_prompt:
call_kwargs["system"] = system_prompt
output_config: Dict[str, Any] = {}
if self.parameters:
if self.parameters.temperature is not None:
call_kwargs["temperature"] = self.parameters.temperature
if self.parameters.top_p is not None:
call_kwargs["top_p"] = self.parameters.top_p
if self.parameters.service_tier is not None:
call_kwargs["service_tier"] = self.parameters.service_tier
if self.parameters.reasoning_effort is not None:
output_config["effort"] = self.parameters.reasoning_effort
if response_model:
schema = response_model.model_json_schema()
forbid_additional_properties(schema)
output_config["format"] = {"type": "json_schema", "schema": schema}
if output_config:
call_kwargs["output_config"] = output_config
prompt_tokens = 0
completion_tokens = 0
full_response = ""
async with self.client.messages.stream(**call_kwargs) as stream:
async for event in stream:
if event.type == "text":
full_response += event.text
await value_streamer.stream_partial(event.text)
final_message = await stream.get_final_message()
if final_message.stop_reason == "refusal":
raise LlmClientException(
f"Anthropic model '{self.model}' refused to generate a response."
)
if final_message.usage:
prompt_tokens = final_message.usage.input_tokens
completion_tokens = final_message.usage.output_tokens
await value_streamer.stream_complete()
duration = time.perf_counter() - start_time
stats = ModelCallStat(
call_type="llm",
model=f"anthropic/{self.model}",
request_data=json.dumps(call_kwargs, default=str),
response_data=full_response,
duration_seconds=duration,
prompt_tokens=prompt_tokens,
completion_tokens=completion_tokens,
total_tokens=prompt_tokens + completion_tokens,
)
await self._send_model_call_stats(stats)
[docs]
def convert_messages(
messages: List[Dict[str, Any]],
) -> Tuple[Optional[str], List[Dict[str, Any]]]:
"""Split chat messages into the system prompt and Anthropic message turns.
The Messages API takes the system prompt as a top-level ``system``
parameter and requires a non-empty ``messages`` list, so a system-only
history (e.g. from :meth:`BaseLlmClient.prompt`) is sent as the single
user turn instead.
"""
system_prompt = None
converted: List[Dict[str, Any]] = []
for msg in messages:
role = msg.get("role")
content = msg.get("content")
if content is None:
continue
if role == "system":
if system_prompt:
system_prompt += "\n" + content
else:
system_prompt = content
continue
anthropic_role = "user" if role == "user" else "assistant"
converted.append({"role": anthropic_role, "content": content})
if not converted:
converted.append({"role": "user", "content": system_prompt or "..."})
system_prompt = None
return system_prompt, converted
[docs]
def forbid_additional_properties(schema: Dict[str, Any]) -> None:
"""
Recursively set 'additionalProperties: false' on all object schemas.
Anthropic structured outputs require closed object schemas.
"""
if not isinstance(schema, dict):
return
if schema.get("type") == "object" or "properties" in schema:
schema["additionalProperties"] = False
# Recursively process nested objects
if "properties" in schema:
for prop_schema in schema["properties"].values():
forbid_additional_properties(prop_schema)
# Handle arrays
if "items" in schema:
forbid_additional_properties(schema["items"])
# Handle allOf, anyOf, oneOf
for key in ["allOf", "anyOf", "oneOf"]:
if key in schema:
for sub_schema in schema[key]:
forbid_additional_properties(sub_schema)
# Handle $defs or definitions (where nested models are stored)
for key in ["$defs", "definitions"]:
if key in schema:
for def_schema in schema[key].values():
forbid_additional_properties(def_schema)