Skip to main content

Long-Term Memory Reference

Long-Term Memory Messaging:​

We would like to introduce Long-Term Memory in Pieces for Developers, powered by our Long-Term Memory Engine (LTME), which enables the world’s first Temporally Grounded Copilot. Long-Term Memory elevates the Pieces Copilot, allowing it to understand your workflow and discuss how, where, and when you work. As with everything we do here at Pieces, this all happens entirely on-device and is available on macOS, Windows, and Linux (coming soon). We truly believe the Pieces Copilot with Long-Term Memory will 10x your productivity. .

Core Benefits​

  1. Reduce context switching and remove the need to search around your desktop to find a piece of information
  2. Reduce the need to add manual context to your copilot as it captures context behind the scenes as you work
  3. Enables natural and intuitive interactions with the Pieces Copilot, allowing you to use it as an assistant that truly understands what you have been working on (prompts below)

Materials to Review and Share​

Privacy and Security​

The Long-Term Memory feature in Pieces enhances the functionality of the Pieces Copilot by utilizing our proprietary Long-Term Memory Engine (LTME). This feature is designed with privacy and efficiency in mind, ensuring that all data processing and storage occur locally on your device.

How Long-Term Memory Works​

  1. On-Device Processing and Storage: All LTME algorithms, processing, and storage take place directly on your device. This ensures that your data remains secure and private, without being transmitted over the internet.
  2. Querying Local Data: When Long-Term Memory is enabled, and you ask a question to the Copilot, the system queries data aggregated from the LTME. This data is processed entirely on your device to find content that is relevant to your query.
  3. Utilizing Retrieval-Augmented Generation (RAG) for Contextual Relevance: The relevant content identified by the LTME is then used as context for the Copilot prompt.
  4. Interaction with Language Models (LLM):
    • Cloud LLM: If you are using a cloud-based LLM, the data identified as relevant is sent to the cloud LLM for processing.
    • Local LLM: If you are using a local LLM, the data remains on your device, ensuring that all processing happens locally without any data leaving your device.

Privacy Recommendations​

For users concerned about privacy, we strongly recommend using a Local LLM with the Pieces Copilot. Options include Mistral, Phi-2, and Llama2, among others. Using a local LLM ensures that all data and processing remain on your device, providing an additional layer of security and privacy.

Performance Note​

Please note that results may vary depending on the selected LLM. Each model has its strengths and capabilities, which can influence the effectiveness of the Long-Term Memory feature.

Use Cases​

Coding​

Debugging and Error Resolution:​

  1. Error Resolution in Specific Functions: Provide specific debugging advice based on the context of the current code and error messages.
    • Prompt: "How can I resolve the issue I'm experiencing in the fetchDataFromAPI function in VS Code?"
  2. Fixing Build Errors:
    • Prompt: "Why is my build failing in the compileProject task in Gradle?"
  3. Error Explanation:
    • Prompt: Explain the error message 'TypeError: undefined is not a function' in my code.

Code Implementation and Enhancement:​

  1. Feature Implementation:
    • Prompt: "Implement the userAuthentication function according to the requirements specified in the Jira ticket."
  2. Refactoring for Maintainability:
    • Prompt: "Refactor the processUserInput function to improve readability and maintainability."

Performance Optimization:​

  1. Optimizing Database Queries:
    • Prompt: "How can I optimize the SQL query in the getUserData function for better performance?"

Code Documentation and Testing​

  1. Generating Documentation: automatically generate documentation comments for functions and classes
    • Prompt: "Generate documentation comments for the calculateMetrics function."
  2. Unit Test Creation: Create unit tests for existing functions
    • Prompt: "Generate unit tests for the validateUserInput function."

Code Comparison and Merging​

  1. Branch Comparison to generate a detailed comparison between two branches in Git:
    • Prompt: "Compare the featureBranch with main and highlight the differences."
  2. Merge Conflict Resolution: Provide guidance on resolving merge conflicts based on the context of the conflicting changes.
    • Prompt: "How should I Resolve merge conflicts between featureBranch and main."

Code Review:​

  1. Generate a summary of changes in a pull request, highlighting key changes and potential issues captured from the pull request.
    • Prompt: Review the pull request number 42 and summarize the changes.

Boilerplate Adaptation:​

  1. Adapt this boilerplate to fit into my existing code.
    • Prompt: "Adapt this boilerplate to fit into my existing code."

Error Recurrence:​

  1. Identify if this error has occurred before in this project and suggest the previous fix.
    • Prompt: "Identify if this error has occurred before in this project and suggest the previous fix."

Research​

  1. Automatically summarize key findings from research papers/websites
    • Prompt: "Summarize the key findings from the research paper and Arxiv on neural network optimization."
    • Prompt: “Extract all the statistical data from this research paper on data mining techniques."
    • Prompt: “Compare the findings of these two research papers on blockchain technology.”
    • Prompt: “Find other research papers that cite this study on machine learning model accuracy.”
  2. Refer to multiple resources to help with validation
  3. What were the main takeaways from the research I did last week on this topic?
    • Prompt: "What were the main takeaways from the research I did last week on this topic?"
  4. What from yesterday’s research can help me here?
    • Prompt: "What from yesterday’s research can help me here?"
  5. What are the steps to implement this tool based on their docs?
    • Prompt: "What are the steps to implement this tool based on their docs?"
  6. Where did I find that Python script I forgot to save in my research yesterday?
    • Prompt: "Where did I find that Python script I forgot to save in my research yesterday?"

Workflow assistance​

Information Retrieval​

  1. Quickly recall any information you've seen, whether it was part of an email, message, or website. This helps you find specific pieces of information without needing to remember where you saw them.
    • Prompt: “What were the specifics of the growth team meeting location in my email?”
    • Prompt: “What were the contact details/email address of the blogger mentioned in the AI blog post on vector search?”
    • Prompt: “Who made that tweet about the Apple Vision Pro review?”
  2. Review multiple resources to help with problem-solving:
    • Prompt: “What is the solution to the problem shared in the support channel around the Pieces OS version upgrade in Windows?”
    • Prompt: “What tasks did I assign to the team in the ClickUp project 'Website Redesign'?”
    • Prompt: “What are the upcoming deadlines for my projects this week?”

Calendar and scheduling​

  1. What are the important meetings in my calendar today?

Conversations/Discussions​

  1. Ask for help related to previous conversations in your workspace: GChat/Slack/Clickup etc.
    • Prompt: What was I talking to Mark about earlier in Google Chat?
    • Prompt: What action items were discussed in the VS Code Channel?
    • Prompt: Based on the given workflow context, have I performed all the necessary release chores in order to release our plugins for VS Code, JetBrains, Obsidian, and JupyterLab?
    • Prompt: “Draft an email to the team summarizing the discussions on the next release in the Slack.”
    • Prompt: "Summarize this active channel in Slack / what my coworker was just explaining."
  2. Write a concise recap of my team meeting based on the transcript.
    • Prompt: "Write a concise recap of my team meeting based on the transcript."

Team Updates​

  1. Use-Case: Summarize recent updates, commits, and communications from team members related to the current sprint
    • Prompt: "What are the latest updates from my team on the current sprint?"

Meeting Preparation:​

  1. Use-Case: Summarize relevant notes, emails, and documents to prepare for upcoming meetings.
    • Prompt: "What are the key points I need to discuss in my meeting with the marketing team?"

Project Collaboration:​

  1. Use-Case: Automatically draft and send a project update to the team based on recent activities and progress.
    • Prompt: "Share the latest project update with the team."

Context Switching Across Various Tools and Improving Productivity​

  1. Can you help me pick up where I left off before lunch?
    • Prompt: "Can you help me pick up where I left off before lunch?"
  2. What was I working on yesterday?
    • Prompt: "What was I working on yesterday?"
  3. Summarize what I did today for my PM.
    • Prompt: "Summarize what I did today for my PM."
  4. What are the main roadblocks for this release based on open GitHub issues?
    • Prompt: "What are the main roadblocks for this release based on open GitHub issues?"
  5. Analyze my recent workflow and suggest ways to reduce context switching.
    • Prompt: "Analyze my recent workflow and suggest ways to reduce context switching."