The basics
Last updated
Last updated
What is Memex good at?
Try asking Memex directly!
Memex has two modes: Build and Code. You can select which one to start a conversation using the toggle under the text input, and you can switch between them at any time during a conversation by toggling it from the top.
Chat
Help with writing
Search the web
Fetch website content
Access files
Create and edit local files
Write code
Run code
Independently iterate based on shell outputs
Perform data science analysis
Generate visualizations
Install libraries and packages
Create and manage environments
We recommend you use chat mode if you want to ensure Memex doesn't build or code anything
Custom instructions are a specific set of directions that you'd like Memex to follow as it helps you on your tasks. You can think of them as extensions of the system prompt. Custom instructions are helpful for directions that you'd like to use across conversations without having to make them part of the prompt each time you start one. Some examples of potential parts of custom instructions:
"Every visualizations should have a transparent background and white labels."
"When transcribing an audio file, default to MLX Whisper or Tiny Whisper. Avoid Base Whisper"
"Keep me updated on the progress by using puns or memes from time to time"
Customer instructions are set under the wrench icon.
Memex can get access to your files via its Build mode. Using its built-in interpreter is able to run code to read, edit and manage them, as well as download specialized libraries for specific file types or use-cases.
For documents or files you expect to use often as part of your project, we recommend you move a copy of the files you'd like a Memex conversation to use to its respective Project. You can open a Project in your Finder by clicking on the project path at the bottom of the UI:
For one-off work or context, you can also drop, copy/paste, or select files using the file picker (image icon):
Technical outputs from Memex are available in drop downs as they are generated to help the conversations stay organized and clean. You can open them at any time to see their content:
One of the key ways that Memex can help you is by running shell commands and processes directly in your computer.
Some of those commands are interactive, and they start and end on their own. They include things like library installations, running scripts with defined start and end, etc.
The other type of commands are known as blocking, because they don't end on their own and require a user to close them or finish. You are most likely to run into blocking commands when launching a app or dashboard.
Memex works with both types of commands, but blocking ones have a specific UI to help you manage them given that a conversation is not able to continue before a blocking command is ended.
Whenever a blocking command is running you'll see three green dots with the words "Running" and a Stop icon next to them. You also won't be able to send any messages when you are in this state to avoid killing them by accident.
You can stop a blocking command by clicking on the Stop icon next to the green dots, or by closing the external process directly (e.g., Python app, streamlit dashboard, etc.)
This is an active area of exploration and we expect to make this experience better soon!
As LLMs continue to evolve and grow, their behavior continues to change as well.
Sometimes, the LLM might feel eager and might start doing more things than you'd like it to do, or do what you asked in a way that doesn't match your expectations. In this cases, and if you'd like to stop it so that it doesn't continue its current exploration vector, you can interrupt it by writing what you'd like it to do. The LLM might need to finish what it's doing before stopping, but being direct and clear will usually help.
(in-line in conversation, no automatic run)
