Nano Apps are tiny applications that can be embedded in Nano Bots. Just as humans can amplify their capabilities with access to tools such as a wrench, a calculator, a screen reader, or internet access, a Nano Bot can enhance its capabilities when equipped with Nano Apps.
Nano Apps can be written in Lua, Fennel, or Clojure. This specific repository has chosen to use Clojure for its Nano Apps.
{index}
Install and set up Nano Bots CLI and Babashka, then:
git clone https://github.com/gbaptista/nano-apps.git
cd nano-apps
bb tasks/generate-cartridge.clj cartridges/template.yml cartridge.yml
# Cartridge successfully generated at cartridge.yml
nb cartridge.yml - repl🤖> What day of the week is it today?
date-and-time {}
{:date-time 2023-12-02T14:19:18-03:00, :timezone America/Sao_Paulo}
Today is Saturday.
🤖> |
You can exit the REPL by typing exit.
{nano-apps}
Just copy and paste any folder inside /apps that resembles what you are trying to do. The structure is:
app.clj: The Nano App source code.specification.yml: The Nano App specification for the Nano Bot.test.clj: Tests for the Nano App.
Additionally:
README.md: A README for humans, not used elsewhere, just to enhance this README.
From a base template, generate a cartridge with all Nano Apps included:
bb tasks/generate-cartridge.clj [CARTRIDGE-TEMPLATE] [OUTPUT-FILE]
# If you don't want to include your private Nano Apps:
bb tasks/generate-cartridge.clj [CARTRIDGE-TEMPLATE] [OUTPUT-FILE] --no-private
bb tasks/generate-cartridge.clj cartridges/template.yml cartridge.yml
# => cartridge.yml
bb tasks/generate-cartridge.clj cartridges/template.yml cartridge.yml --no-private
# => cartridge.ymlAfter generating a cartridge, considering that you have Nano Bots CLI, you can try it right away:
nb cartridge.yml - repl
🤖> what time is it?
date-and-time {}
{:date-time 2023-12-02T14:10:05Z, :timezone UTC}
The current time in UTC is 14:10:05 on December 2, 2023.
🤖> |
You can exit the REPL by typing exit.
Running all tests:
bb tasks/run-tests.cljRunning individual tests:
bb apps/simple-calculator/test.cljRunning all tests, excluding private Nano Apps:
bb tasks/run-tests.clj --no-private Update the template.md file and then:
bb tasks/generate-readme.cljTrick for automatically updating the README.md when template.md changes:
sudo pacman -S inotify-tools # Arch / Manjaro
sudo apt-get install inotify-tools # Debian / Ubuntu / Raspberry Pi OS
sudo dnf install inotify-tools # Fedora / CentOS / RHEL
while inotifywait -e modify -e create template.md apps/*/README.md; do bb tasks/generate-readme.clj; doneTrick for Markdown Live Preview:
pip install -U markdown_live_preview
mlp README.md -p 8076bb tasks/cljfmt-fix.cljYou can create private Nano Apps that will not be committed to the repository by using the pattern user@app:
apps/your-name@your-app/app.clj
The same applies to private cartridge templates:
cartridges/your-name@your-cartridge-template.yml
You can exclude your private Nano Apps from cartridge generation with:
bb tasks/generate-cartridge.clj cartridges/template.yml cartridge.yml --no-privateRunning all tests, excluding private Nano Apps:
bb tasks/run-tests.clj --no-private Nano Apps are designed as part of Nano Bots, so it is important to familiarize yourself with the Nano Bots' specification and principles: Nano Bots Specification
A Nano App is only as good as the AI provider's capability to leverage it. Therefore, as a core principle, everything you have learned about Prompt Engineering should be considered and applied to the development of Nano Apps, especially when writing their specifications:
Sometimes you may feel like, "this is too complex," or "it will not be useful," or "why would it use this," etc. You may be surprised by the creative ways Nano Apps are used by models.
For a concrete example, the Media Player Control has commands that you may think aren't worth providing. But, if you spend some time playing around, you'll have many "wow" moments about how the model uses it to achieve what you are asking for.
So, give it room to be creative and explore, don't hold back on functionality just because you feel unsure.
Although LLMs are becoming increasingly capable of handling substantial amounts of data, we need to consider the following when it comes to tokens:
- Excessively large outputs may lead to a decrease in model accuracy during reasoning about content.
- Generating large outputs incurs costs since payment is typically on a per-token basis.
- An excessive number of tokens may overload a model's capacity, leading to errors.
Avoid providing unnecessarily lengthy outputs; seize the opportunity to make your outputs concise. Be careful when building Nano Apps that query databases or run commands, as they can produce unexpectedly lengthy outputs.
When using Nano Apps as a proxy for popular software, ensure that the output is faithful to the actual software. Models reason better when the output is exactly what it would expect from the program.
Example, if your Nano App is a proxy for GNU Octave software, the expected output for operations would be:
ans = 6
It may be tempting to shorten it to return only 6 instead of ans = 6. Don't do that; keep the original output, as it was what the model learned to expect during its training.
Keep errors concise (Concise Outputs) and helpfull (Prompt Engineering).
When using Nano Apps as a proxy for popular software, prioritize the sharing of the original error output from the software (Output Fidelity). Some software generates lengthy error stack traces. In this scenario, it is best to provide alternative custom short messages or restrict the output size without sacrificing necessary detail. Balance Output Fidelity with Concise Outputs.
Returning () as the output of a function can cause OpenAI models to request that the function run again, potentially leading to an infinite loop.
- Nano Bots Specification
- Nano Bots CLI
- Nano Bots API
- Nano Bots for Sublime Text
- Nano Bots for Visual Studio Code
This is an experimental, early-stage project. Nano Apps may be dangerous, so be careful with what you try to build; they could execute destructive actions on your computer. Also, be mindful of your budget: Ensure you monitor and budget whatever provider you are using. Nano Apps may produce unexpectedly lengthy content or infinite/too-long loops that could cause your costs to skyrocket.
This software is distributed under the MIT License, which includes a disclaimer of warranty. Furthermore, the authors assume no responsibility for any damage or costs that may arise from the use of this experimental, early-stage project. Use Nano Apps at your own risk.