syndarix/backend/app/services/context/adapters/openai.py

"""
OpenAI Model Adapter.

Provides OpenAI-specific context formatting using markdown
which GPT models understand well.
"""

from typing import Any

from ..types import BaseContext, ContextType
from .base import ModelAdapter


class OpenAIAdapter(ModelAdapter):
    """
    OpenAI-specific context formatting adapter.

    GPT models work well with markdown formatting,
    so we use headers and structured markdown for clarity.

    Features:
    - Markdown headers for each context type
    - Bulleted lists for document sources
    - Bold role labels for conversations
    - Code blocks for tool outputs
    """

    MODEL_PATTERNS: list[str] = ["gpt", "openai", "o1", "o3"]

    def format(
        self,
        contexts: list[BaseContext],
        **kwargs: Any,
    ) -> str:
        """
        Format contexts for OpenAI models.

        Uses markdown formatting for structured content.

        Args:
            contexts: List of contexts to format
            **kwargs: Additional formatting options

        Returns:
            Markdown-structured context string
        """
        if not contexts:
            return ""

        by_type = self.group_by_type(contexts)
        parts: list[str] = []

        for ct in self.get_type_order():
            if ct in by_type:
                formatted = self.format_type(by_type[ct], ct, **kwargs)
                if formatted:
                    parts.append(formatted)

        return self.get_separator().join(parts)

    def format_type(
        self,
        contexts: list[BaseContext],
        context_type: ContextType,
        **kwargs: Any,
    ) -> str:
        """
        Format contexts of a specific type for OpenAI.

        Args:
            contexts: List of contexts of the same type
            context_type: The type of contexts
            **kwargs: Additional formatting options

        Returns:
            Markdown-formatted string for this context type
        """
        if not contexts:
            return ""

        if context_type == ContextType.SYSTEM:
            return self._format_system(contexts)
        elif context_type == ContextType.TASK:
            return self._format_task(contexts)
        elif context_type == ContextType.KNOWLEDGE:
            return self._format_knowledge(contexts)
        elif context_type == ContextType.CONVERSATION:
            return self._format_conversation(contexts)
        elif context_type == ContextType.TOOL:
            return self._format_tool(contexts)

        return "\n".join(c.content for c in contexts)

    def _format_system(self, contexts: list[BaseContext]) -> str:
        """Format system contexts."""
        content = "\n\n".join(c.content for c in contexts)
        return content

    def _format_task(self, contexts: list[BaseContext]) -> str:
        """Format task contexts."""
        content = "\n\n".join(c.content for c in contexts)
        return f"## Current Task\n\n{content}"

    def _format_knowledge(self, contexts: list[BaseContext]) -> str:
        """
        Format knowledge contexts as structured documents.

        Each knowledge context becomes a section with source attribution.
        """
        parts = ["## Reference Documents\n"]

        for ctx in contexts:
            source = ctx.source
            score = ctx.metadata.get("score", ctx.metadata.get("relevance_score", ""))

            if score:
                parts.append(f"### Source: {source} (relevance: {score})\n")
            else:
                parts.append(f"### Source: {source}\n")

            parts.append(ctx.content)
            parts.append("")

        return "\n".join(parts)

    def _format_conversation(self, contexts: list[BaseContext]) -> str:
        """
        Format conversation contexts as message history.

        Uses bold role labels for clear turn delineation.
        """
        parts = []

        for ctx in contexts:
            role = ctx.metadata.get("role", "user").upper()
            parts.append(f"**{role}**: {ctx.content}")

        return "\n\n".join(parts)

    def _format_tool(self, contexts: list[BaseContext]) -> str:
        """
        Format tool contexts as tool results.

        Each tool result is in a code block with the tool name.
        """
        parts = ["## Recent Tool Results\n"]

        for ctx in contexts:
            tool_name = ctx.metadata.get("tool_name", "unknown")
            status = ctx.metadata.get("status", "")

            if status:
                parts.append(f"### Tool: {tool_name} ({status})\n")
            else:
                parts.append(f"### Tool: {tool_name}\n")

            parts.append(f"```\n{ctx.content}\n```")
            parts.append("")

        return "\n".join(parts)