66063abb8c
When an LLM returns content: null on a plain assistant message (no tool_calls), the null gets saved to session history and causes permanent 400 errors on every subsequent request. - Sanitize None content on plain assistant messages to "(empty)" in _sanitize_empty_content(), matching the existing empty-string handling - Skip persisting error responses (finish_reason="error") to the message history in _run_agent_loop(), preventing poison loops Closes #1303
119 lines
3.9 KiB
Python
119 lines
3.9 KiB
Python
"""Base LLM provider interface."""
|
|
|
|
from abc import ABC, abstractmethod
|
|
from dataclasses import dataclass, field
|
|
from typing import Any
|
|
|
|
|
|
@dataclass
|
|
class ToolCallRequest:
|
|
"""A tool call request from the LLM."""
|
|
id: str
|
|
name: str
|
|
arguments: dict[str, Any]
|
|
|
|
|
|
@dataclass
|
|
class LLMResponse:
|
|
"""Response from an LLM provider."""
|
|
content: str | None
|
|
tool_calls: list[ToolCallRequest] = field(default_factory=list)
|
|
finish_reason: str = "stop"
|
|
usage: dict[str, int] = field(default_factory=dict)
|
|
reasoning_content: str | None = None # Kimi, DeepSeek-R1 etc.
|
|
|
|
@property
|
|
def has_tool_calls(self) -> bool:
|
|
"""Check if response contains tool calls."""
|
|
return len(self.tool_calls) > 0
|
|
|
|
|
|
class LLMProvider(ABC):
|
|
"""
|
|
Abstract base class for LLM providers.
|
|
|
|
Implementations should handle the specifics of each provider's API
|
|
while maintaining a consistent interface.
|
|
"""
|
|
|
|
def __init__(self, api_key: str | None = None, api_base: str | None = None):
|
|
self.api_key = api_key
|
|
self.api_base = api_base
|
|
|
|
@staticmethod
|
|
def _sanitize_empty_content(messages: list[dict[str, Any]]) -> list[dict[str, Any]]:
|
|
"""Replace empty text content that causes provider 400 errors.
|
|
|
|
Empty content can appear when MCP tools return nothing. Most providers
|
|
reject empty-string content or empty text blocks in list content.
|
|
"""
|
|
result: list[dict[str, Any]] = []
|
|
for msg in messages:
|
|
content = msg.get("content")
|
|
|
|
# None content on a plain assistant message (no tool_calls) crashes
|
|
# providers with "invalid message content type: <nil>".
|
|
if content is None and msg.get("role") == "assistant" and not msg.get("tool_calls"):
|
|
clean = dict(msg)
|
|
clean["content"] = "(empty)"
|
|
result.append(clean)
|
|
continue
|
|
|
|
if isinstance(content, str) and not content:
|
|
clean = dict(msg)
|
|
clean["content"] = None if (msg.get("role") == "assistant" and msg.get("tool_calls")) else "(empty)"
|
|
result.append(clean)
|
|
continue
|
|
|
|
if isinstance(content, list):
|
|
filtered = [
|
|
item for item in content
|
|
if not (
|
|
isinstance(item, dict)
|
|
and item.get("type") in ("text", "input_text", "output_text")
|
|
and not item.get("text")
|
|
)
|
|
]
|
|
if len(filtered) != len(content):
|
|
clean = dict(msg)
|
|
if filtered:
|
|
clean["content"] = filtered
|
|
elif msg.get("role") == "assistant" and msg.get("tool_calls"):
|
|
clean["content"] = None
|
|
else:
|
|
clean["content"] = "(empty)"
|
|
result.append(clean)
|
|
continue
|
|
|
|
result.append(msg)
|
|
return result
|
|
|
|
@abstractmethod
|
|
async def chat(
|
|
self,
|
|
messages: list[dict[str, Any]],
|
|
tools: list[dict[str, Any]] | None = None,
|
|
model: str | None = None,
|
|
max_tokens: int = 4096,
|
|
temperature: float = 0.7,
|
|
) -> LLMResponse:
|
|
"""
|
|
Send a chat completion request.
|
|
|
|
Args:
|
|
messages: List of message dicts with 'role' and 'content'.
|
|
tools: Optional list of tool definitions.
|
|
model: Model identifier (provider-specific).
|
|
max_tokens: Maximum tokens in response.
|
|
temperature: Sampling temperature.
|
|
|
|
Returns:
|
|
LLMResponse with content and/or tool calls.
|
|
"""
|
|
pass
|
|
|
|
@abstractmethod
|
|
def get_default_model(self) -> str:
|
|
"""Get the default model for this provider."""
|
|
pass
|