SecondBrain is a CLI tool to access and filter data from external data sources such as Salesforce, ZenDesk, Slack, Gong and others, and then query that data using an LLM.

This tool uses Retrieval Augmented Generation (RAG) techniques to pull in relevant data from external sources, filter it based on keywords, and then pass that data to a Large Language Model (LLM) for analysis.

This is useful for generating reports, summaries, and insights from multiple and otherwise disconnected data sources.

Building

Clone the git repository:

git clone https://github.com/mcasperson/SecondBrain.git

Build the project using Maven:

./mvnw clean package -DskipTests

LLM Backend

The easiest way to get started is to use Ollama as the LLM backend. Install Ollama from https://ollama.com and then pull the qwen3:30b-a3b model:

ollama pull qwen3:30b-a3b

You are free to use a variety of models, but I have found the qwen3 models to be some of the best.

Directory Scan Example

The directory scan tools allows you to scan a directory of files for keywords, extract relevant text, and then ask questions about the data found. This is useful when you need to analyze a collection of documents, such as PDFs, Word documents, or text files.

Run the DirectoryScan tool against the PDF files in the samples directory, looking for keywords "AI", "Kubernetes", and "K8s", and then ask a question about the data found. Replace the replaceme values with your Azure AI Foundry API key and endpoint URL, and also replace Phi-4 with the model you wish to use:

java \
    "-Dsb.llm.client=azure" \
    "-Dsb.azurellm.apikey=replaceme" \
    "-Dsb.azurellm.model=Phi-4" \
    "-Dsb.azurellm.url=https://replaceme.services.ai.azure.com/models/chat/completions?api-version=2024-05-01-preview" \
    "-Dsb.tools.force=DirectoryScan" \
    "-Dsb.directoryscan.directory=samples" \
    "-Dsb.directoryscan.keywords=AI,Kubernetes,K8s" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "What percentage of AI developers are cloud native?"

This command runs the example with Ollama:

ollama pull qwen3:30b-a3b
java \
    "-Dsb.llm.client=ollama" \
    "-Dsb.ollama.model=qwen3:30b-a3b" \
    "-Dsb.ollama.contextwindow=30000" \
    "-Dsb.tools.force=DirectoryScan" \
    "-Dsb.directoryscan.directory=samples" \
    "-Dsb.directoryscan.keywords=AI,Kubernetes,K8s" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "What percentage of AI developers are cloud native?"

GitHub Diff Example

java \
    "-Dsb.llm.client=azure" \
    "-Dsb.azurellm.apikey=replaceme" \
    "-Dsb.azurellm.url=https://replaceme.services.ai.azure.com/models/chat/completions?api-version=2024-05-01-preview" \
    "-Dsb.azurellm.model=Phi-4" \
    "-Dsb.tools.force=GitHubDiffs" \
    "-Dsb.github.owner=mcasperson" \
    "-Dsb.github.repo=SecondBrain" \
    "-Dsb.github.days=30" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "Write a 3 paragraph summary of the changes made to the SecondBrain project in the last 30 days."

java \
    "-Dsb.llm.client=ollama" \
    "-Dsb.ollama.model=qwen3:30b-a3b" \
    "-Dsb.ollama.contextwindow=30000" \
    "-Dsb.tools.force=GitHubDiffs" \
    "-Dsb.github.owner=mcasperson" \
    "-Dsb.github.repo=SecondBrain" \
    "-Dsb.github.days=30" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "Write a 3 paragraph summary of the changes made to the SecondBrain project in the last 30 days."

YouTube Transcript Example

This example downloads the transcripts from videos from a YouTube playlist trims the transcripts to the keywords, and then generates a summary of the AI related news from those videos.

You must supply an YouTube API key with access to the YouTube Data API v3, and set the playlist ID and keywords to filter the transcripts.

Note that this is a fairly slow process as the YouTube API is heavily rate limited. We process 2 videos a minute, so this command may take up to 5 minutes to complete:

java \
    "-Dsb.llm.client=azure" \
    "-Dsb.azurellm.apikey=replaceme" \
    "-Dsb.azurellm.url=https://replaceme.services.ai.azure.com/models/chat/completions?api-version=2024-05-01-preview" \
    "-Dsb.azurellm.model=Phi-4" \
    "-Dsb.tools.force=YoutubePlaylist" \
    "-Dsb.google.apiKey=apikeygoeshere" \
    "-Dsb.youtube.playlistId=PLlrxD0HtieHgFYS4DKbJ_xCYNE94ZLJjj" \
    "-Dsb.youtube.maxvideos=10" \
    "-Dsb.youtube.keywords=AI,LLM,MCP,Agent" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "Write a 3 paragraph summary of the AI related news from the YouTube videos."

This is the same command using Ollama:

ollama pull qwen3:30b-a3b
java \
    "-Dsb.llm.client=ollama" \
    "-Dsb.ollama.model=qwen3:30b-a3b" \
    "-Dsb.tools.force=YoutubePlaylist" \
    "-Dsb.google.apiKey=apikeygoeshere" \
    "-Dsb.youtube.playlistId=PLlrxD0HtieHgFYS4DKbJ_xCYNE94ZLJjj" \
    "-Dsb.youtube.maxvideos=10" \
    "-Dsb.youtube.keywords=AI,LLM,MCP,Agent" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "Write a 3 paragraph summary of the AI related news from the YouTube videos."

ZenDesk Example

The ZenDeskOrganization tool connects to a ZenDesk instance, retrieves support tickets from the specified number of days, and passes the relevant data to the LLM for analysis.

Here is the command using Azure AI Foundry. Replace the replaceme values with your Azure AI Foundry API key and endpoint URL, and also replace Phi-4 with the model you wish to use:

java \
    "-Dsb.llm.client=azure" \
    "-Dsb.azurellm.apikey=replaceme" \
    "-Dsb.azurellm.url=https://replaceme.services.ai.azure.com/models/chat/completions?api-version=2024-05-01-preview" \
    "-Dsb.azurellm.model=Phi-4" \
    "-Dsb.tools.force=ZenDeskOrganization" \
    "-Dsb.zendesk.url=https://replaceme.zendesk.com" \
    "-Dsb.zendesk.user=replaceme@replaceme.org" \
    "-Dsb.zendesk.accesstoken=replaceme" \
    "-Dsb.zendesk.days=30" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "Summarize the top customer support issues from the last 30 days."

Here is the command using Ollama:

ollama pull qwen3:30b-a3b
java \
    "-Dsb.llm.client=ollama" \
    "-Dsb.ollama.model=qwen3:30b-a3b" \
    "-Dsb.tools.force=ZenDeskOrganization" \
    "-Dsb.zendesk.url=https://replaceme.zendesk.com" \
    "-Dsb.zendesk.user=replaceme@replaceme.org" \
    "-Dsb.zendesk.accesstoken=replaceme" \
    "-Dsb.zendesk.days=30" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "Summarize the top customer support issues from the last 30 days."

Gong Example

The Gong tool connects to a Gong instance, retrieves sales calls from the specified number of days, and passes the relevant data to the LLM for analysis.

Here is the command using Azure AI Foundry. Replace the replaceme values with your Azure AI Foundry API key and endpoint URL, and also replace Phi-4 with the model you wish to use:

java \
    "-Dsb.llm.client=azure" \
    "-Dsb.azurellm.apikey=replaceme" \
    "-Dsb.azurellm.url=https://replaceme.services.ai.azure.com/models/chat/completions?api-version=2024-05-01-preview" \
    "-Dsb.azurellm.model=Phi-4" \
    "-Dsb.tools.force=Gong" \
    "-Dsb.gong.url=https://api.gong.io" \
    "-Dsb.gong.accessKey=replaceme" \
    "-Dsb.gong.accessSecretKey=replaceme" \
    "-Dsb.gong.days=30" \
    "-Dsb.gong.company=replaceme" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "Summarize the common topics discussed in the last 30 days."

ollama pull qwen3:30b-a3b
java \
    "-Dsb.llm.client=ollama" \
    "-Dsb.ollama.model=qwen3:30b-a3b" \
    "-Dsb.tools.force=Gong" \
    "-Dsb.gong.url=https://api.gong.io" \
    "-Dsb.gong.accessKey=replaceme" \
    "-Dsb.gong.accessSecretKey=replaceme" \
    "-Dsb.gong.days=30" \
    "-Dsb.gong.company=replaceme" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "Summarize the common topics discussed in the last 30 days."

Meta Tool Example

The Meta tool combines the results of multiple tools into a single report. This is useful when you need to aggregate data from multiple sources and then ask questions about the combined data.

In this example we combine the YoutubePlaylist and DirectoryScan tools to generate a summary of AI related news from both YouTube videos and PDF reports in the samples directory.

Note the use of the sb.meta.postInferenceHooks setting to enable the MetaDataHook post inference processing step. We then set the contextMetaField and contextMetaPrompt parameters to add additional context to the report. The results of the prompts are captured in the file called meta.json which can be reviewed after the command completes.

This is useful when the results of this operation are consumed by another process. For example, you could build a report that indicated which documents mentioned Python or Kubernetes based on the results of the context meta prompts:

ollama pull qwen3:30b-a3bs
java \
     "-Dsb.llm.client=ollama" \
     "-Dsb.ollama.model=qwen3:30b-a3b" \
     "-Dsb.ollama.contextwindow=30000" \
     "-Dsb.tools.force=Meta" \
     "-Dsb.meta.toolNames=YoutubePlaylist,DirectoryScan" \
     "-Dsb.meta.keywords=AI,LLM,MCP,Agent,K8s" \
     "-Dsb.meta.postInferenceHooks=MetaDataHook" \
     "-Dsb.metadatahook.metareport=meta.json" \
     "-Dsb.metadatahook.contextMetaField1=Python" \
     "-Dsb.metadatahook.contextMetaPrompt1=Does the content describe anything related to the Python programming language?" \
     "-Dsb.metadatahook.contextMetaField2=Kubernetes" \
     "-Dsb.metadatahook.contextMetaPrompt2=Does the content describe anything related to Kubernetes?" \
     "-Dsb.youtube.playlistId=PLlrxD0HtieHgFYS4DKbJ_xCYNE94ZLJjj" \
     "-Dsb.youtube.maxvideos=10" \
     "-Dsb.directoryscan.directory=samples" \
     -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
     "Write a 3 paragraph summary of the AI related news from the YouTube videos and PDF reports."

Extra logging

To enable more verbose logging, you must create a file like the following:

# Edd the console handler
handlers=java.util.logging.ConsoleHandler
# Enabling FINE level logging
java.util.logging.ConsoleHandler.level=FINE
# Infrastructure classes interact with external systems
secondbrain.infrastructure.level=FINE
# Tools are responsible for integrating the external systems with the LLM
secondbrain.domain.tools.level=FINE

Then reference that file using the java.util.logging.config.file system property:

java \
    "-Djava.util.logging.config.file=logging.properties" \
    "-Dsb.llm.client=azure" \
    "-Dsb.azurellm.apikey=replaceme" \
    "-Dsb.azurellm.model=Phi-4" \
    "-Dsb.azurellm.url=https://replaceme.services.ai.azure.com/models/chat/completions?api-version=2024-05-01-preview" \
    "-Dsb.tools.force=DirectoryScan" \
    "-Dsb.directoryscan.directory=samples" \
    "-Dsb.directoryscan.keywords=AI,Kubernetes,K8s" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "What percentage of AI deployments use Kubernetes or K8s for orchestration?"

Building your own tools

You can build your own tools by implementing the Tool interface found in the secondbrain-core module.

See the HelloWorld tool for an example of how to build your own tool.

java \
    "-Dsb.llm.client=ollama" \
    "-Dsb.ollama.model=qwen3:30b-a3b" \
    "-Dsb.tools.force=HelloWorld" \
    "-Dsb.helloworld.message=AI Fans" \
    -jar cli/target/secondbrain-cli-1.0-SNAPSHOT.jar \
    "The prompt is not used with the HelloWorld tool." 

Project Structure

The project is split into modules:

• secondbrain-core which contains shared utilities and interfaces used by all other modules.
• secondbrain-service which orchestrates the function calling with the LLM.
• secondbrain-tools which contains the tools that interact with external data sources.
• secondbrain-web which is a web interface for interacting with the service.
• secondbrain-cli which is a CLI tool for interacting with the service.

Awards

SecondBrain was awarded second place in the Payara Hackathon - Generative AI on Jakarta EE!