The world of GenAI agents is fragmented, with frameworks like LangGraph, AWS Bedrock, and Semantic Kernel each operating in their own silos. This lack of interoperability creates hurdles for developers and enterprises needing to integrate agents across different platforms, APIs, or legacy systems.
AgenticDB emerges as the solution to unify this scattered landscape. It's a database specifically designed to store and manage GenAI agent and application manifests. With AgenticDB, you can effortlessly add, search, invoke, and rate agents, all from a centralized repository. Advanced features like Docker-based agent management and remote execution expand its versatility, making it adaptable for a wide array of GenAI workflows.
By streamlining the way we handle agents, AgenticDB not only saves time but also fosters innovation by making it easier to find and utilize the right agent for any task. It's the bridge that connects disparate frameworks, enabling seamless collaboration and accelerating progress in the GenAI ecosystem.
With AgenticDB, you can:
- Store and manage agent manifests.
- Search for agents using similarity search.
- Invoke agents via local and remote execution models.
- Rate agents based on performance and interaction results.
To start the AgenticDB server locally, run the following command:
python server.pyThe API will be available at http://127.0.0.1:8000.
Here's an updated version of the Delete All Collections section in the README, reflecting the actual JSON response format from the provided Python code.
You can add an agent to the database using the following curl command. This example sends a JSON request to add a new agent called code-gen-chart-agent.
curl -X POST "http://127.0.0.1:8000/agents" \
-H "Content-Type: application/json" \
-H "Accept: application/json" \
-d '[
{
"metadata": {
"name": "code-gen-chart-agent",
"namespace": "agents",
"description": "Requests for a javascript code generator to plot a chart with supplied free-form data."
},
"spec": {
"type": "agent",
"lifecycle": "dev",
"owner": "admin@company.com",
"access_level": "Public",
"category": "Travel Agent",
"url": "https://api.example.com/code-gen-chart-agent/agent",
"parameters": {
"type": "object",
"properties": {
"message": {
"type": "string",
"description": "The request message for the chart code generator agent."
},
"thread": {
"type": "string",
"description": "The id to separate parallel message threads."
}
},
"required": ["message", "thread"],
"additionalProperties": false
},
"output": {
"type": "string",
"description": "The result of the request, including any generated image location."
}
}
}
]'This will add the agent manifest to AgenticDB, making it available for future searches and invocations.
To search for agents based on a natural language query, use the following curl command. This example searches for agents related to "Natural Language."
curl -G "http://127.0.0.1:8000/agents" \
-H "Accept: application/json" \
--data-urlencode "query=Which agents can book travel?"AgenticDB will perform a similarity search and return a list of matching agents based on the query.
You can also add an agentic application to AgenticDB using a similar curl command. Here’s an example for adding a "Stock Price Charting Application" that tracks stock prices and generates charts.
curl -X POST "http://127.0.0.1:8000/applications" \
-H "Content-Type: application/json" \
-H "Accept: application/json" \
-d '[
{
"metadata": {
"name": "Stock Price Charting Application",
"namespace": "production",
"description": "Provides access to daily open, high, low, close stock prices over time and the ability to generate charts for the requested data"
},
"spec": {
"type": "application",
"lifecycle": "dev",
"owner": "admin@company.com",
"access_level": "PUBLIC",
"category": "Finance",
"setup": {
"compose": {
"compose_url": "https://ipfs.filebase.io/ipfs/somehash",
"run_command": "gunzip docker-compose.yml.gz && docker compose -f ./docker-compose.yml up -d"
}
},
"url": "http://localhost:3000/agent",
"method": "POST",
"example": "http://localhost:3000/agent 'Content-Type':'application/json' {'input':'what was the Nvidia close price on August 22nd 2024', 'thread':'nvidia'}",
"parameters": {
"type": "object",
"properties": {
"input": {
"type": "string",
"description": "natural language request for stock price data and charting of the data as required"
},
"thread": {
"type": "string",
"description": "thread context id for the request"
}
},
"required": ["input"],
"additionalProperties": false
},
"output": {
"type": "object",
"properties": {
"content": {
"type": "string",
"description": "the natural language response and the Final Answer to the request with a chart location if requested"
}
},
"description": "the answer to the request"
}
}
}
]'You can delete all agent, application, and rating collections from AgenticDB with a single DELETE request. This operation will attempt to remove the stored collections: agents, applications, and ratings. Each collection deletion is handled separately, and the results will reflect whether each collection was successfully deleted.
curl -X DELETE "http://127.0.0.1:8000/collections"The response will provide the status of each deletion attempt in the form of a JSON object. If a collection is successfully deleted, the value will be 0. If there was an error, the value will contain the corresponding error message.
{
"agents": 0,
"applications": 0,
"ratings": "Failed to delete ratings collection: some_error_message"
}- agents:
0indicates theagentscollection was successfully deleted. - applications:
0indicates theapplicationscollection was successfully deleted. - ratings: Contains an error message if the deletion failed (replace
"some_error_message"with the actual error encountered).
After interacting with an agent, you can submit a rating for the agent using this curl command. Replace placeholder_agent_id and placeholder_some_id with the actual agent and rating IDs.
curl -X POST "http://127.0.0.1:8000/ratings" \
-H "Content-Type: application/json" \
-d '{
"ratings": {
"agent_id": "placeholder_agent_id",
"id": "placeholder_some_id",
"data": {
"score": 4
}
}
}'You can retrieve agent ratings using this curl command. Replace <ratings_id> with the actual ratings ID.
curl -X GET "http://127.0.0.1:8000/ratings?ratings_id=<ratings_id>"AgenticDB will return the score and feedback associated with the provided ratings ID.