Exo – the electron app that’s helping to build itself
Hi HN folks! I’m the creator of Exo and wanted to share my progress with the HN community. Exo is an electron app that uses a chat interface and open Ai to write code directly to your file system/repos. Exo is a work in progress - its very new and has many rough edges. I was hoping to get some early feedback from HN folks and hopefully some early users. Once you create an account the universal electron installer link is on the dashboard. Here is a quick explainer video: https://www.loom.com/share/5f34499ccfb54bfdae32ee50f454b365 Exo started as a thin wrapper around…
What it does
In the maker’s words, at launch
Hi HN folks! I’m the creator of Exo and wanted to share my progress with the HN community. Exo is an electron app that uses a chat interface and open Ai to write code directly to your file system/repos. Exo is a work in progress - its very new and has many rough edges. I was hoping to get some early feedback from HN folks and hopefully some early users. Once you create an account the universal electron installer link is on the dashboard. Here is a quick explainer video: https://www.loom.com/share/5f34499ccfb54bfdae32ee50f454b365 Exo started as a thin wrapper around my file system. I built a bunch of functions that let me use open Ai to write/refactor code directly in my repos. I wanted to build some simple tools to help increase my programming productivity. At one point I realized this app could be useful to other engineers so I started using it to build itself… hence the clickbait title. Don’t hate me :). Here’s how it works: All of the chat functionality (including message classification and processing) uses GPT-3.5 turbo so it’s pretty fast. There is a bunch of logic that uses OpenAI to classify and adapt the incoming user messages to route them to the correct place (search, write code, ask for clarification, etc). I’ve experimented with an “agent"(https://python.langchain.com/en/latest/modules/agents/agents...) model for this but it didn’t work as well as the classification layer. Exo uses GPT-4 to write code. I’ve found a pretty big difference in code quality and “correctness” between GPT-3.5 and GPT-4. It doesn’t fall back to 3.5 but if 4 stays flakey I may have to build a fall back mechanism. You can index a (small) repo or subset of a repo (components section for example) for search. I recommend starting with a directory with 10-50 files to start. This will take 2-3 min to index. After it runs you can search based on the code functionality. For example “find the component that renders the login page…”. The indexing process uses tree sitter to parse the files. It then uses GPT-3.5 turbo to summarize what the code does. It then embeds the code itself with the explanation. When you search, it embeds your search phrase and matches based on the code explanation. I’m using supabase for the db. I’m using the pgvector postgres add on to store the embeddings. This is very useful as the code explanation and all the metadata sits right next to the embedding so returning the search list is super fast. Its also simple to render given it doesn’t need to pull data from multiple places. I plan on adding the functionality to add your own OpenAI api code soon. For now the app uses my personal OpenAI account. So you get to use GPT, including GPT-4, for free :). Don’t go crazy. GPT-4 can be pretty flakey. There are some days when 50% or more api requests get throttled. I’ve added a basic throttling function but the OpenAI api still throws a 429 with nontrivial consistency. I suspect if a bunch of people are using Exo at the same time most of the calls will fail making it nonfunctional. I’ll post an update and likely cap signups if that happens. I’d love feedback on the Exo app itself. Mostly if you’d use something like this to increase your programming throughput. The static site was just thrown together and needs a bunch of work so don’t judge me on that :). Also a lot styling/layout in the app is pretty poor. I’ve basically focused exclusively on functionality at the expense of looking nice. I’ve been posting videos of my progress and how it works to my Twitter feed which you can find here: https://twitter.com/KevinGrassi If you have any feedback or questions you can email me at my hn handle at gmail. I’d love to hear from you! Thanks for reading this far!
Does the same job
all alternatives →
- HXHacker XP – Hacker News styled as a Windows XP Outlook email client2022 · hackerxp.com · ▲136
Hello HN, sharing my weekend project. I'm a fan of retro GUIs and fan of HN, so I built an alternative HN front end that resembles a Window XP desktop with an old Outlook email client. Archive and github links if you get 500 errors: https://web.archive.org/web/20220122190454/https://hackerxp.... https://github.com/assemblylanguage/hacker-xp
- EAEno – A lightning fast, user-friendly YAML/TOML alternative2018 · ▲102
eno [1] - A modern plaintext language w/ libraries [2] for JavaScript, Python, Ruby & soon more! We migrated a big relational research database to a file-based solution - requirements were: - Super fast and easy editability for users - Highest performance for parsing/validating >10K documents on every user change. Our trials with YAML/TOML showed us that we wanted something both faster [3][4] and easier [4], something tailored for file-based content management ... and after months of research & development it's now publicly available (under MIT license) for everyone! Last but…
- HSHelium - Simple web automation2013 · heliumhq.com · ▲48
Hi everyone, I'm working for a startup called BugFree Software and would love to hear your feedback on a new product we are launching today. Helium is a library that wraps around Selenium to simplify web automation. It does away with many of the technicalities involved with web scripting. For example: Here is a Selenium script. Can you guess what it does? >>> ff = Firefox() ... >>> text_area = ff.find_element_by_id("u_0_1q") >>> text_area.send_keys("Hello World!") >>> button = ff.find_element_by_class_name("_42g-") >>> button.click() Here is the same script rewritten using Helium: >>>…
- HAHain – Alternative to Alfred on Windows, Built with Electron2016 · github.com · ▲12
- KEKotlin/Elm/Electron Tech Demo2017 · ▲7
Another HN Desktop client: https://github.com/spennihana/HackerDesktop Last weekend user westoncb posted[1] an app built in kotlin/swing[2]. A comment there mentioned that an electron app might be an interesting alternative, and here is the result of this last week's hackings. It uses the firebase API underneath and elm/electron to interact with kotlin via the spark-kotlin micro rest framework. The icns were made with Adobe Illustrator. The dmg unpacks to a whopping 250MB (since it packages the JRE in addition to all of the usual electron junk). video link:…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, April 2023
the whole month →
- G4
Hi HN, Today we’re launching GPT-4 answers on Phind.com, a developer-focused search engine that uses generative AI to browse the web and answer technical questions, complete with code examples and detailed explanations. Unlike vanilla GPT-4, Phind feeds in relevant websites and technical documentation, reducing the model’s hallucination and keeping it up-to-date. To use it, simply enable the “Expert” toggle before doing a search. GPT-4 is making a night-and-day difference in terms of answer quality. For a question like “How can I RLHF a LLaMa model”, Phind in Expert mode delivers a…
AI · 2023 · phind.com
