[backend] agent tool metadata db

[scott-cohere]

• create new table
• create tool metadata entries on agent creation
• new crud endpoints for the table
• unit test

AI Description

This PR introduces the AgentToolMetadata model and its associated CRUD operations, tests, and API endpoints.

Code Changes:

• Added a new AgentToolMetadata model in src/backend/database_models/agent_tool_metadata.py to represent agent tool metadata.
• Modified src/backend/routers/agent.py to include endpoints for creating, listing, updating, and deleting agent tool metadata.
• Added new functions in src/backend/crud/agent_tool_metadata.py for CRUD operations on AgentToolMetadata.
• Updated src/backend/chat/custom/custom.py to include agent_id in the call_tools function.
• Added tests for AgentToolMetadata CRUD operations in src/backend/tests/routers/test_agent.py and src/backend/tests/crud/test_agent_tool_metadata.py.
• Added a new AgentToolMetadataFactory in src/backend/tests/factories/agent_tool_metadata.py for generating test data.
• Updated imports in multiple files to include the new AgentToolMetadata model and related functions.

Summary:
The main purpose of this PR is to add support for managing agent tool metadata, including creating, updating, listing, and deleting metadata associated with specific agents. This enhancement provides a structured way to store and retrieve metadata related to agent tools, improving the flexibility and maintainability of the system.

[giannis2two]

As discussed, alternative design to be more generalized for more tools

class AgentToolArtifact(Base):
    __tablename__ = "agent_tool_artifact"

    user_id: Mapped[str] = mapped_column(Text, nullable=False)
    agent_id: Mapped[str] = mapped_column(
        ForeignKey("agents.id", ondelete="CASCADE"), nullable=False
    )
    tool_name: Mapped[str] = mapped_column(Text, nullable=False)
    artifact_id: Mapped[str] = mapped_column(Text, nullable=False)
    type: Mapped[str] = mapped_column(Text, nullable=False)

    __table_args__ = (
        UniqueConstraint(
            "user_id", "agent_id", "tool_name", "artifact_id", name="_user_agent_tool_name_artifact_id_uc"
        ),
    )

Then you can do the following

• Create a new entry for each folder/file/url
  • One to many (agent to artifacts)
• With type you can define an arbitrary string that the client (backend) can use to conditionally parse
  • If folder and google_drive then I know what to do
  • If url and web_search then I know what to do

[scott-cohere]

@giannis2two
Slight deviation from proposed payload for passing in the artifacts, it is now this:

        "tools_metadata": [
            {
                "tool_name": ToolName.Google_Drive,
                "artifacts": ["folder1", "folder2"],
                "type": "folder_ids",
            },
            {
                "tool_name": ToolName.Google_Drive,
                "artifacts": ["file1", "file2"],
                "type": "file_ids",
            },
        ],
   

This plays better with the DB schema and a lot easier to parse. I don't imagine it would be a huge issue to parse like this on the client side.

[giannis2two]

High level looks great! 🥳

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 96.28253% with 10 lines in your changes missing coverage. Please review.

Please upload report for BASE (main@14f5b38). Learn more about missing BASE report.

Files	Patch %	Lines
src/backend/routers/agent.py	81.63%	9 Missing ⚠️
src/backend/alembic/versions/2d766d30a3d2_.py	91.66%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #309   +/-   ##
=======================================
  Coverage        ?   86.13%           
=======================================
  Files           ?      172           
  Lines           ?     6421           
  Branches        ?        0           
=======================================
  Hits            ?     5531           
  Misses          ?      890           
  Partials        ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.