mindroom 0.0.0__py3-none-any.whl → 0.1.0__py3-none-any.whl

mindroom/agents.py

@@ -0,0 +1,248 @@
+"""Agent loader that reads agent configurations from YAML file."""
+
+from __future__ import annotations
+
+from datetime import datetime
+from typing import TYPE_CHECKING
+from zoneinfo import ZoneInfo
+
+from agno.agent import Agent
+from agno.storage.sqlite import SqliteStorage
+
+from . import agent_prompts
+from . import tools as _tools_module  # noqa: F401
+from .constants import ROUTER_AGENT_NAME, SESSIONS_DIR
+from .logging_config import get_logger
+from .tools_metadata import get_tool_by_name
+
+if TYPE_CHECKING:
+    from .config import Config
+
+logger = get_logger(__name__)
+
+# Maximum length for instruction descriptions to include in agent summary
+MAX_INSTRUCTION_LENGTH = 100
+
+
+def get_datetime_context(timezone_str: str) -> str:
+    """Generate current date and time context for the agent.
+
+    Args:
+        timezone_str: Timezone string (e.g., 'America/New_York', 'UTC')
+
+    Returns:
+        Formatted string with current date and time information
+
+    """
+    tz = ZoneInfo(timezone_str)
+    now = datetime.now(tz)
+
+    # Format the datetime in a clear, readable way
+    date_str = now.strftime("%A, %B %d, %Y")
+    time_str = now.strftime("%H:%M %Z")  # 24-hour format
+
+    return f"""## Current Date and Time
+Today is {date_str}.
+The current time is {time_str} ({timezone_str} timezone).
+
+"""
+
+
+# Rich prompt mapping - agents that use detailed prompts instead of simple roles
+RICH_PROMPTS = {
+    "code": agent_prompts.CODE_AGENT_PROMPT,
+    "research": agent_prompts.RESEARCH_AGENT_PROMPT,
+    "calculator": agent_prompts.CALCULATOR_AGENT_PROMPT,
+    "general": agent_prompts.GENERAL_AGENT_PROMPT,
+    "shell": agent_prompts.SHELL_AGENT_PROMPT,
+    "summary": agent_prompts.SUMMARY_AGENT_PROMPT,
+    "finance": agent_prompts.FINANCE_AGENT_PROMPT,
+    "news": agent_prompts.NEWS_AGENT_PROMPT,
+    "data_analyst": agent_prompts.DATA_ANALYST_AGENT_PROMPT,
+}
+
+
+def create_agent(agent_name: str, config: Config) -> Agent:
+    """Create an agent instance from configuration.
+
+    Args:
+        agent_name: Name of the agent to create
+        config: Application configuration
+
+    Returns:
+        Configured Agent instance
+
+    Raises:
+        ValueError: If agent_name is not found in configuration
+
+    """
+    from .ai import get_model_instance  # noqa: PLC0415
+
+    # Use passed config (config_path is deprecated)
+    agent_config = config.get_agent(agent_name)
+    defaults = config.defaults
+
+    # Create tools
+    tools: list = []  # Use list type to satisfy Agent's parameter type
+    for tool_name in agent_config.tools:
+        try:
+            tool = get_tool_by_name(tool_name)
+            tools.append(tool)
+        except ValueError as e:
+            logger.warning(f"Could not load tool '{tool_name}' for agent '{agent_name}': {e}")
+
+    SESSIONS_DIR.mkdir(parents=True, exist_ok=True)
+    storage = SqliteStorage(table_name=f"{agent_name}_sessions", db_file=str(SESSIONS_DIR / f"{agent_name}.db"))
+
+    # Get model config for identity context
+    model_name = agent_config.model or "default"
+    if model_name in config.models:
+        model_config = config.models[model_name]
+        model_provider = model_config.provider.title()  # Capitalize provider name
+        model_id = model_config.id
+    else:
+        # Fallback if model not found
+        model_provider = "AI"
+        model_id = model_name
+
+    # Add identity context to all agents using the unified template
+    identity_context = agent_prompts.AGENT_IDENTITY_CONTEXT.format(
+        display_name=agent_config.display_name,
+        agent_name=agent_name,
+        model_provider=model_provider,
+        model_id=model_id,
+    )
+
+    # Add current date and time context with user's configured timezone
+    datetime_context = get_datetime_context(config.timezone)
+
+    # Combine identity and datetime contexts
+    full_context = identity_context + datetime_context
+
+    # Use rich prompt if available, otherwise use YAML config
+    if agent_name in RICH_PROMPTS:
+        logger.info(f"Using rich prompt for agent: {agent_name}")
+        # Prepend full context to the rich prompt
+        role = full_context + RICH_PROMPTS[agent_name]
+        instructions = []  # Instructions are in the rich prompt
+    else:
+        logger.info(f"Using YAML config for agent: {agent_name}")
+        # For YAML agents, prepend full context to role and keep original instructions
+        role = full_context + agent_config.role
+        instructions = agent_config.instructions
+
+    # Create agent with defaults applied
+    model = get_model_instance(config, agent_config.model)
+    logger.info(f"Creating agent '{agent_name}' with model: {model.__class__.__name__}(id={model.id})")
+
+    instructions.append(agent_prompts.INTERACTIVE_QUESTION_PROMPT)
+
+    agent = Agent(
+        name=agent_config.display_name,
+        role=role,
+        model=model,
+        tools=tools,
+        instructions=instructions,
+        storage=storage,
+        add_history_to_messages=agent_config.add_history_to_messages
+        if agent_config.add_history_to_messages is not None
+        else defaults.add_history_to_messages,
+        num_history_runs=agent_config.num_history_runs or defaults.num_history_runs,
+        markdown=agent_config.markdown if agent_config.markdown is not None else defaults.markdown,
+    )
+    logger.info(f"Created agent '{agent_name}' ({agent_config.display_name}) with {len(tools)} tools")
+
+    return agent
+
+
+def describe_agent(agent_name: str, config: Config) -> str:
+    """Generate a description of an agent or team based on its configuration.
+
+    Args:
+        agent_name: Name of the agent or team to describe
+        config: Application configuration
+
+    Returns:
+        Human-readable description of the agent or team
+
+    """
+    # Handle built-in router agent
+    if agent_name == ROUTER_AGENT_NAME:
+        return (
+            "router\n"
+            "  - Route messages to the most appropriate agent based on context and expertise.\n"
+            "  - Analyzes incoming messages and determines which agent is best suited to respond."
+        )
+
+    # Check if it's a team
+    if agent_name in config.teams:
+        team_config = config.teams[agent_name]
+        parts = [f"{agent_name}"]
+        if team_config.role:
+            parts.append(f"- {team_config.role}")
+        parts.append(f"- Team of agents: {', '.join(team_config.agents)}")
+        parts.append(f"- Collaboration mode: {team_config.mode}")
+        return "\n  ".join(parts)
+
+    # Check if agent exists
+    if agent_name not in config.agents:
+        return f"{agent_name}: Unknown agent or team"
+
+    agent_config = config.agents[agent_name]
+
+    # Start with agent name (not display name, for routing consistency)
+    parts = [f"{agent_name}"]
+    if agent_config.role:
+        parts.append(f"- {agent_config.role}")
+
+    # Add tools if any
+    if agent_config.tools:
+        tool_list = ", ".join(agent_config.tools)
+        parts.append(f"- Tools: {tool_list}")
+
+    # Add key instructions if any
+    if agent_config.instructions:
+        # Take first instruction as it's usually the most descriptive
+        first_instruction = agent_config.instructions[0]
+        if len(first_instruction) < MAX_INSTRUCTION_LENGTH:  # Only include if reasonably short
+            parts.append(f"- {first_instruction}")
+
+    return "\n  ".join(parts)
+
+
+def get_agent_ids_for_room(room_key: str, config: Config) -> list[str]:
+    """Get all agent Matrix IDs assigned to a specific room."""
+    # Always include the router agent
+    agent_ids = [config.ids[ROUTER_AGENT_NAME].full_id]
+
+    # Add agents from config
+    for agent_name, agent_cfg in config.agents.items():
+        if room_key in agent_cfg.rooms:
+            agent_ids.append(config.ids[agent_name].full_id)
+    return agent_ids
+
+
+def get_rooms_for_entity(entity_name: str, config: Config) -> list[str]:
+    """Get the list of room aliases that an entity (agent/team) should be in.
+
+    Args:
+        entity_name: Name of the agent or team
+        config: Configuration object
+
+    Returns:
+        List of room aliases the entity should be in
+
+    """
+    # TeamBot check (teams)
+    if entity_name in config.teams:
+        return config.teams[entity_name].rooms
+
+    # Router agent special case - gets all rooms
+    if entity_name == ROUTER_AGENT_NAME:
+        return list(config.get_all_configured_rooms())
+
+    # Regular agents
+    if entity_name in config.agents:
+        return config.agents[entity_name].rooms
+
+    return []

mindroom/ai.py

@@ -0,0 +1,421 @@
+"""AI integration module for MindRoom agents and memory management."""
+
+from __future__ import annotations
+
+import functools
+import os
+from typing import TYPE_CHECKING, Any
+
+import diskcache
+from agno.models.anthropic import Claude
+from agno.models.cerebras import Cerebras
+from agno.models.google import Gemini
+from agno.models.ollama import Ollama
+from agno.models.openai import OpenAIChat
+from agno.models.openrouter import OpenRouter
+from agno.run.response import (
+    RunResponse,
+    RunResponseContentEvent,
+    ToolCallCompletedEvent,
+    ToolCallStartedEvent,
+)
+
+from .agents import create_agent
+from .constants import ENABLE_AI_CACHE
+from .credentials_sync import get_api_key_for_provider, get_ollama_host
+from .error_handling import get_user_friendly_error_message
+from .logging_config import get_logger
+from .memory import build_memory_enhanced_prompt
+
+if TYPE_CHECKING:
+    from collections.abc import AsyncIterator
+    from pathlib import Path
+
+    from agno.agent import Agent
+    from agno.models.base import Model
+
+    from .config import Config, ModelConfig
+
+logger = get_logger(__name__)
+
+
+def _extract_response_content(response: RunResponse) -> str:
+    response_parts = []
+
+    # Add main content if present
+    if response.content:
+        response_parts.append(response.content)
+
+    # Add formatted tool calls if present (similar to agno's print_response)
+    # Only add if there are actual tool calls to display
+    if response.formatted_tool_calls and any(response.formatted_tool_calls):
+        tool_calls_section = "\n\n**Tool Calls:**"
+        for tool_call in response.formatted_tool_calls:
+            tool_calls_section += f"\n• {tool_call}"
+        response_parts.append(tool_calls_section)
+
+    return "\n".join(response_parts) if response_parts else ""
+
+
+def _format_tool_started_message(event: ToolCallStartedEvent) -> str:
+    if not event.tool:
+        return ""
+
+    tool_name = event.tool.tool_name if event.tool.tool_name else "tool"
+    tool_args = event.tool.tool_args if event.tool.tool_args else {}
+
+    # Format similar to agno's formatted_tool_calls
+    if tool_args:
+        args_str = ", ".join(f"{k}={v}" for k, v in tool_args.items())
+        msg = f"\n\n🔧 **Tool Call:** `{tool_name}({args_str})`\n"
+    else:
+        msg = f"\n\n🔧 **Tool Call:** `{tool_name}()`\n"
+
+    return msg
+
+
+def _format_tool_completed_message(event: ToolCallCompletedEvent) -> str:
+    if not event.tool:
+        return ""
+
+    tool_name = event.tool.tool_name if event.tool.tool_name else "tool"
+
+    # Check both event.content and tool.result for the output
+    result = event.content or (event.tool.result if event.tool else None)
+
+    if result:
+        # Format the result nicely
+        return f"✅ **`{tool_name}` result:**\n{result}\n\n"
+
+    return f"✅ **`{tool_name}`** completed\n\n"
+
+
+@functools.cache
+def get_cache(storage_path: Path) -> diskcache.Cache | None:
+    """Get or create a cache instance for the given storage path."""
+    return diskcache.Cache(storage_path / ".ai_cache") if ENABLE_AI_CACHE else None
+
+
+def _set_api_key_env_var(provider: str) -> None:
+    """Set environment variable for a provider from CredentialsManager.
+
+    Since we sync from .env to CredentialsManager on startup,
+    this will always use the latest keys from .env.
+
+    Args:
+        provider: Provider name (e.g., 'openai', 'anthropic')
+
+    """
+    # Map provider names to environment variable names
+    env_vars = {
+        "openai": "OPENAI_API_KEY",
+        "anthropic": "ANTHROPIC_API_KEY",
+        "openrouter": "OPENROUTER_API_KEY",
+        "gemini": "GOOGLE_API_KEY",
+        "google": "GOOGLE_API_KEY",
+        "cerebras": "CEREBRAS_API_KEY",
+        "deepseek": "DEEPSEEK_API_KEY",
+        "groq": "GROQ_API_KEY",
+    }
+
+    if provider not in env_vars:
+        return
+
+    # Get API key from CredentialsManager (which has been synced from .env)
+    api_key = get_api_key_for_provider(provider)
+
+    # Set environment variable if key exists
+    if api_key:
+        os.environ[env_vars[provider]] = api_key
+        logger.debug(f"Set {env_vars[provider]} from CredentialsManager")
+
+
+def _create_model_for_provider(provider: str, model_id: str, model_config: ModelConfig, extra_kwargs: dict) -> Model:
+    """Create a model instance for a specific provider.
+
+    Args:
+        provider: The AI provider name
+        model_id: The model identifier
+        model_config: The model configuration object
+        extra_kwargs: Additional keyword arguments for the model
+
+    Returns:
+        Instantiated model for the provider
+
+    Raises:
+        ValueError: If provider not supported
+
+    """
+    if provider == "ollama":
+        # Priority: model config > env/CredentialsManager > default
+        # This allows per-model host configuration in config.yaml
+        host = model_config.host or get_ollama_host() or "http://localhost:11434"
+        logger.debug(f"Using Ollama host: {host}")
+        return Ollama(id=model_id, host=host, **extra_kwargs)
+    if provider == "openai":
+        return OpenAIChat(id=model_id, **extra_kwargs)
+    if provider == "anthropic":
+        return Claude(id=model_id, **extra_kwargs)
+    if provider == "openrouter":
+        # OpenRouter needs the API key passed explicitly because it captures
+        # the environment variable at import time, not at instantiation time
+        api_key = get_api_key_for_provider(provider)
+        if not api_key:
+            logger.warning("No OpenRouter API key found in environment or CredentialsManager")
+        return OpenRouter(id=model_id, api_key=api_key, **extra_kwargs)
+    if provider in ("gemini", "google"):
+        return Gemini(id=model_id, **extra_kwargs)
+    if provider == "cerebras":
+        return Cerebras(id=model_id, **extra_kwargs)
+
+    msg = f"Unsupported AI provider: {provider}"
+    raise ValueError(msg)
+
+
+def get_model_instance(config: Config, model_name: str = "default") -> Model:
+    """Get a model instance from config.yaml.
+
+    Args:
+        config: Application configuration
+        model_name: Name of the model configuration to use (default: "default")
+
+    Returns:
+        Instantiated model
+
+    Raises:
+        ValueError: If model not found or provider not supported
+
+    """
+    if model_name not in config.models:
+        available = ", ".join(sorted(config.models.keys()))
+        msg = f"Unknown model: {model_name}. Available models: {available}"
+        raise ValueError(msg)
+
+    model_config = config.models[model_name]
+    provider = model_config.provider
+    model_id = model_config.id
+
+    logger.info("Using AI model", model=model_name, provider=provider, id=model_id)
+
+    # Set environment variable from CredentialsManager for Agno to use
+    _set_api_key_env_var(provider)
+
+    # Get extra kwargs if specified
+    extra_kwargs = model_config.extra_kwargs or {}
+
+    return _create_model_for_provider(provider, model_id, model_config, extra_kwargs)
+
+
+def _build_full_prompt(prompt: str, thread_history: list[dict[str, Any]] | None = None) -> str:
+    """Build full prompt with thread history context."""
+    if not thread_history:
+        return prompt
+
+    context = "Previous conversation in this thread:\n"
+    for msg in thread_history:
+        context += f"{msg['sender']}: {msg['body']}\n"
+    context += "\nCurrent message:\n"
+    return context + prompt
+
+
+def _build_cache_key(agent: Agent, full_prompt: str, session_id: str) -> str:
+    model = agent.model
+    assert model is not None
+    return f"{agent.name}:{model.__class__.__name__}:{model.id}:{full_prompt}:{session_id}"
+
+
+async def _cached_agent_run(
+    agent: Agent,
+    full_prompt: str,
+    session_id: str,
+    agent_name: str,
+    storage_path: Path,
+) -> RunResponse:
+    """Cached wrapper for agent.arun() calls."""
+    cache = get_cache(storage_path)
+    if cache is None:
+        return await agent.arun(full_prompt, session_id=session_id)  # type: ignore[no-any-return]
+
+    model = agent.model
+    assert model is not None
+    cache_key = _build_cache_key(agent, full_prompt, session_id)
+    cached_result = cache.get(cache_key)
+    if cached_result is not None:
+        logger.info("Cache hit", agent=agent_name)
+        return cached_result  # type: ignore[no-any-return]
+
+    response = await agent.arun(full_prompt, session_id=session_id)
+
+    cache.set(cache_key, response)
+    logger.info("Response cached", agent=agent_name)
+
+    return response  # type: ignore[no-any-return]
+
+
+async def _prepare_agent_and_prompt(
+    agent_name: str,
+    prompt: str,
+    storage_path: Path,
+    room_id: str | None,
+    config: Config,
+    thread_history: list[dict[str, Any]] | None = None,
+) -> tuple[Agent, str]:
+    """Prepare agent and full prompt for AI processing.
+
+    Returns:
+        Tuple of (agent, full_prompt, session_id)
+
+    """
+    enhanced_prompt = await build_memory_enhanced_prompt(prompt, agent_name, storage_path, config, room_id)
+    full_prompt = _build_full_prompt(enhanced_prompt, thread_history)
+    logger.info("Preparing agent and prompt", agent=agent_name, full_prompt=full_prompt)
+    agent = create_agent(agent_name, config)
+    return agent, full_prompt
+
+
+async def ai_response(
+    agent_name: str,
+    prompt: str,
+    session_id: str,
+    storage_path: Path,
+    config: Config,
+    thread_history: list[dict[str, Any]] | None = None,
+    room_id: str | None = None,
+) -> str:
+    """Generates a response using the specified agno Agent with memory integration.
+
+    Args:
+        agent_name: Name of the agent to use
+        prompt: User prompt
+        session_id: Session ID for conversation tracking
+        storage_path: Path for storing agent data
+        config: Application configuration
+        thread_history: Optional thread history
+        room_id: Optional room ID for room memory access
+
+    Returns:
+        Agent response string
+
+    """
+    logger.info("AI request", agent=agent_name)
+
+    # Prepare agent and prompt - this can fail if agent creation fails (e.g., missing API key)
+    try:
+        agent, full_prompt = await _prepare_agent_and_prompt(
+            agent_name,
+            prompt,
+            storage_path,
+            room_id,
+            config,
+            thread_history,
+        )
+    except Exception as e:
+        logger.exception("Error preparing agent", agent=agent_name)
+        return get_user_friendly_error_message(e, agent_name)
+
+    # Execute the AI call - this can fail for network, rate limits, etc.
+    try:
+        response = await _cached_agent_run(agent, full_prompt, session_id, agent_name, storage_path)
+    except Exception as e:
+        logger.exception("Error generating AI response", agent=agent_name)
+        return get_user_friendly_error_message(e, agent_name)
+
+    # Extract response content - this shouldn't fail
+    return _extract_response_content(response)
+
+
+async def stream_agent_response(  # noqa: C901, PLR0912
+    agent_name: str,
+    prompt: str,
+    session_id: str,
+    storage_path: Path,
+    config: Config,
+    thread_history: list[dict[str, Any]] | None = None,
+    room_id: str | None = None,
+) -> AsyncIterator[str]:
+    """Generate streaming AI response using Agno's streaming API.
+
+    Checks cache first - if found, yields the cached response immediately.
+    Otherwise streams the new response and caches it.
+
+    Args:
+        agent_name: Name of the agent to use
+        prompt: User prompt
+        session_id: Session ID for conversation tracking
+        storage_path: Path for storing agent data
+        config: Application configuration
+        thread_history: Optional thread history
+        room_id: Optional room ID for room memory access
+
+    Yields:
+        Chunks of the AI response as they become available
+
+    """
+    logger.info("AI streaming request", agent=agent_name)
+
+    # Prepare agent and prompt - this can fail if agent creation fails
+    try:
+        agent, full_prompt = await _prepare_agent_and_prompt(
+            agent_name,
+            prompt,
+            storage_path,
+            room_id,
+            config,
+            thread_history,
+        )
+    except Exception as e:
+        logger.exception("Error preparing agent for streaming", agent=agent_name)
+        yield get_user_friendly_error_message(e, agent_name)
+        return
+
+    # Check cache (this shouldn't fail)
+    cache = get_cache(storage_path)
+    if cache is not None:
+        model = agent.model
+        assert model is not None
+        cache_key = _build_cache_key(agent, full_prompt, session_id)
+        cached_result = cache.get(cache_key)
+        if cached_result is not None:
+            logger.info("Cache hit", agent=agent_name)
+            response_text = cached_result.content or ""
+            yield response_text
+            return
+
+    full_response = ""
+
+    # Execute the streaming AI call - this can fail for network, rate limits, etc.
+    try:
+        stream_generator = await agent.arun(full_prompt, session_id=session_id, stream=True)
+    except Exception as e:
+        logger.exception("Error starting streaming AI response")
+        yield get_user_friendly_error_message(e, agent_name)
+        return
+
+    # Process the stream events
+    try:
+        async for event in stream_generator:
+            if isinstance(event, RunResponseContentEvent) and event.content:
+                chunk_text = str(event.content)
+                full_response += chunk_text
+                yield chunk_text
+            elif isinstance(event, ToolCallStartedEvent):
+                tool_msg = _format_tool_started_message(event)
+                if tool_msg:
+                    full_response += tool_msg
+                    yield tool_msg
+            elif isinstance(event, ToolCallCompletedEvent):
+                result_msg = _format_tool_completed_message(event)
+                if result_msg:
+                    full_response += result_msg
+                    yield result_msg
+            else:
+                logger.warning(f"Unhandled event type: {type(event).__name__} - {event}")
+    except Exception as e:
+        logger.exception("Error during streaming AI response")
+        yield get_user_friendly_error_message(e, agent_name)
+        return
+
+    if cache is not None and full_response:
+        cached_response = RunResponse(content=full_response)
+        cache.set(cache_key, cached_response)
+        logger.info("Response cached", agent=agent_name)

mindroom/api/__init__.py

@@ -0,0 +1 @@
+"""Backend initialization for the widget module."""