Show HN: It took me 5() months to build a Plausible alternative After months of using Google Analytics I realized only about 50% of people accepted my cookie-popup. I had months of incorrect data for my website. I started looking for alternatives and eventually found Plausible, which is great (and open-source). Problem is, I didn't feel like paying 9$ a month to see the amount of visitors on a website i didn't even earn anything on, it was just a hobby project. Eventually I started making my own web analytics. Which actually isn't that hard. It took me about a month of working on my spare time every now and then. Being GDPR compliant basically means to not save any personal identifiers. At first I thought it would be easy since something like a public IP adress can't count as a personal identifier right? I was very wrong. How it works: When a user visited my website I saved the IP and Header for 24 hours. Then if they visited again I checked the combination of IP and Header against the ones saved in my DB. If they were the same I simply added 1 view to my data. If they weren't the same I added 1 unique daily user and 1 view. That's in short how it works. A few weeks later I realized if I had this problem then other would also have it. So I started working on Simplytics.dev. I had to do a lot of new stuff and re-build my code from the ground-up twice. Small things like OAuth was completely new to me and took up a lot of time. But eventually I got here and just launched something that with the knowledge I have today wouldn't even take a third of the time recreating today. It's my first real "Launch" and it feels really good finally creating something AND publishing it. Instead of a montly fee I opted to make it a pay-once service. Right now it's priced at 49$ but I'll see how it works out. If you got any questions on how it works Id love to answer them. November 25, 2024 at 09:05PM
Show HN: Built This in 3 Hours Using Bolt (No React Knowledge) It's blowing my mind that I could build an app from scratch in 3 hours and deploy it with the click of a button - without writing a single line of code myself CacheNotes is a browser-based note-taking app that saves your notes, threads, and AI conversations securely in your browser's local storage 100% free. No login required. You get, - simple, minimal note taking app in your browser - no login required - you can connect your claude api key for the ai integration - visualise notes and twitter threads Check it out! Would love some feedback :) https://ift.tt/rksmBzn November 30, 2024 at 12:39AM
Show HN: I built an extension to contact Airbnb hosts direcly Hey HN, Over the years, I’ve found myself frustrated with the extra fees Airbnb. While they provide a lot of convenience, the service fees often stack up to a point where they overshadow any potential savings. I started wondering if there was a way to connect directly with property hosts and skip the middleman entirely. That curiosity led me to build getaway.direct, a free Chrome extension that helps travelers save money. It works like this: 1. You browse listings on Airbnb. 2. The extension scans for direct booking links, host websites, or social media profiles where you can reach out to the host directly. 3. It shows those results instantly, so you can compare prices, avoid service fees, and book smarter. The main idea is to provide transparency. A lot of hosts already have their own websites but rely on platforms like Airbnb for visibility. This tool helps surface those direct options, which can often save you 10-20% per booking. I’d love to hear your thoughts: Would something like this be helpful to you? Also, I’m curious to get feedback on ways to improve the tool—whether it’s adding more integrations, improving usability, or something else entirely. Thanks for reading! Let me know what you think. https://ift.tt/w8B3s01 November 29, 2024 at 04:48PM
Show HN: A word guessing game based on text vector embeddings and cos-similarity Try to find the secret word that computer holds, by guessing and getting feedback in form of how similar your guess to the secret is. The fewer attempts the better. There is also a hint and a give-up button. Thank you, please give it a try ) https://ift.tt/i4PyTg5 November 29, 2024 at 04:41AM
Show HN: MyDuck Server – Supercharge MySQL and Postgres Analytics with DuckDB Hello HN! We're excited to announce MyDuck Server, an open-source project that seamlessly integrates the analytical power of DuckDB with your existing MySQL & Postgres databases. *Backstory* Currently, there are no fully satisfactory open-source OLAP solutions for MySQL & Postgres. In the MySQL ecosystem, HeatWave offers close integration, but it's a proprietary, commercial product from Oracle. The Postgres community has seen promising DuckDB-based extensions emerge, including the official pg_duckdb. However, extensions can introduce isolation concerns in mission-critical environments. Consequently, many organizations resort to setting up complex and costly data movement pipelines using tools like Debezium, Flink, or other commercial solutions to replicate data from MySQL & Postgres to OLAP systems (e.g., Snowflake, BigQuery, ClickHouse) or Lakehouses (e.g., Delta Lake + Spark). This approach introduces significant operational overhead and expense. Another emerging strategy is the zero-ETL approach, increasingly advocated by cloud providers. This model simplifies data integration by allowing the cloud provider to manage ETL pipelines, while necessitating reliance on specific cloud ecosystems and services. *Key features* MyDuck Server offers a real-time analytical replica that leverages DuckDB's native columnar storage and processing capabilities. It operates as a separate server, ensuring isolation and minimizing impact on your primary database. Key features include: - Easy Zero-ETL: Built-in real-time replication from MySQL & Postgres with no complex pipelines to manage. It feels like a standard MySQL replica or Postgres standby. With the Docker image, passing a connection string is enough. - MySQL & Postgres Protocol Compatibility: We take this seriously and are working to make this project integrate well with the existing ecosystem around MySQL & Postgres. Currently, it is already possible to connect to MyDuck with standard MySQL & PostgreSQL clients in many programming languages. - HTAP Support: A standard database proxy can be deployed in front of a MySQL/Postgres primary and its MyDuck replica to route write operations to the primary and read operations to the replica. It just works. - DuckDB SQL & Columnar I/O over Postgres Protocol: It's unnecessary to restrict ourselves to MySQL/Postgres's SQL expressiveness and row-oriented data transfer. The Postgres port accepts all DuckDB-valid SQL queries, and you can retrieve query results in columnar format via `COPY (SELECT ...) TO STDOUT (FORMAT parquet/arrow)`. - Standalone Mode: It does not need to be run as a replica. It can also act as a primary server that brings DuckDB into server mode and accepts updates from multiple connections, breaking DuckDB's single-process limitation. *Relevant Previous HN Threads* - pg_duckdb [1] ( https://ift.tt/nOXkat5 ) is the official Postgres extension for DuckDB. It uses DuckDB as an execution engine to accelerate analytical queries by scanning Postgres tables directly. - pg_mooncake [2] ( https://ift.tt/pvxWRtE ) is a Postgres extension that adds columnstore tables for PG. It uses pg_duckdb under the hood but stores data in Lakehouse formats (Iceberg & Delta Lake). - BemiDB [3] ( https://ift.tt/gnJTQsN ) is also a DuckDB-based Postgres replica. Unlike us, they focus on storing data in Lakehouse format. We believe MyDuck Server offers a compelling solution for those seeking high-performance analytics on their MySQL & Postgres data without the complexities and costs of traditional approaches. We're eager to hear your feedback and answer any questions you might have. Let me know what you think! [0] https://ift.tt/ouGtn2f [1] https://ift.tt/FIwW6ka [2] https://ift.tt/QnrTHqI [3] https://ift.tt/zsWqjLU https://ift.tt/ouGtn2f November 28, 2024 at 08:20PM
Show HN: AirPrint Bridge: Enable AirPrint for Non-AirPrint Printers on macOS Lightweight, open-source, and fully automated, AirPrint Bridge uses macOS's built-in tools to bridge the gap without relying on external software. Perfect for reviving your trusty old printer! https://ift.tt/YpkIlv9 November 27, 2024 at 11:41PM
Show HN: AutoPiP – Safari extension for automatic Picture-in-Picture mode AutoPiP is a Safari extension that automatically enables Picture-in-Picture mode when switching tabs and disables it when returning to the video tab. No clicks needed - just seamless multitasking while watching videos. Key features: • Automatic PiP activation when switching tabs • Smart detection for active videos only • Seamless disable on tab return • Zero configuration needed GitHub: https://ift.tt/amH3pYb https://ift.tt/amH3pYb November 27, 2024 at 04:06PM
Show HN: Clean Your Mac with a Script I wanted to clean old temporary files and caches from my macOS with a script instead of using a shady paid app, so I created a simple script for that. Pull requests are very welcome for other unused files to clean up storage space! https://ift.tt/Bkp0Qbo November 27, 2024 at 03:19AM
Show HN: Neon-PostgREST – Build a Supabase-Like API/SDK DevX over Neon Hey folks - one thing I find a bit frustrating using non-supabase Postgres providers is the lack of ability to perform DB queries via an SDK/API. I decided to build a quick sample of how to combine PostgREST + Neon to recreate this dev experience. https://ift.tt/fvKG3iX November 25, 2024 at 11:12PM
Show HN: QuackHouse, Database in the browser, using WASM and DuckDB I'm building a privacy focused analytics tool, using WebAssembly and DuckDB. You can upload your files (CSV, JSON and Parquet), and interact with them as where they a SQL Server. Your data never leaves your computer, however I do track page views and visitors using Plausible. The next steps are to add forecasting and segmentation, as well as some data visualisation capabilities. I would love to hear your opinion. All code for the repo is available here: https://ift.tt/oxULP0Z https://ift.tt/178GQUO November 25, 2024 at 02:42AM
Show HN: Over 600 CSS Animations to Code Hey HN, We’re excited to share Gradienty's CSS Animation Generator designed to make web animations intuitive and accessible for developers and designers at any level. Whether you’re new to CSS animations or a seasoned pro looking to save time, Gradienty equips you with the tools to create beautiful animations with zero coding headaches. Key Features: 1. 600+ Pre-Built Animations: From subtle hover effects to complex keyframe sequences, all categorized for easy navigation. 2. Visual Editor with Live Preview: See your animations in action as you tweak timing, easing, delay, and iterations. 3. Responsive Design Previews: Test animations across layouts for desktop and mobile compatibility. 4. Multiple Preview Objects: Visualize animations on text, buttons, boxes, circles, and more. 5. One-Click Code Export: Generate production-ready CSS with proper vendor prefixes, ready to drop into your project. 6. Zero-Dependency Animations: Works flawlessly across all modern browsers. Why We Built This: As developers, we often found animations to be either overly complex to implement manually or limited by pre-made libraries. Gradienty bridges this gap by offering both flexibility and ease of use, helping you create animations that look and perform great—without sacrificing development time. Who It’s For: Beginners: Experiment with animations visually without writing a single line of code. Designers: Focus on creativity while leaving the technical aspects to the generator. Developers: Save time with ready-to-use animations that can be customized and exported instantly. What’s Next: We’re working on adding community features like user-created animation libraries, animation presets for specific design systems, and integration guides for popular frameworks. We’d love to hear your feedback! Check it out here: Gradienty Let us know what you think or if there are features you’d love to see! https://ift.tt/8lLz5PI November 24, 2024 at 01:07AM
Show HN: EnvCloak: lightweight, env file encryption – ready for CICD pipelines When someone here told me to focus on something more useful than reinventing the wheel. So. EnvCloak, a lightweight and simple tool for securely managing sensitive environment files. The design focuses on simplicity - just a few intuitive commands using the Click Python library. I assume seamless integration with CI/CD workflows. The aim is to provide a streamlined solution without the need for clunky tools. If you're tired of complex configurations or bloated alternatives, this might be worth a look! I would appreciate any feedback, feature ideas or input on how to improve this solution. I'd love to hear your thoughts! Regards! https://ift.tt/fhkcz5E November 23, 2024 at 11:29PM
Show HN: AI bot that automatically processes unstructured documents Hi HN! We’re excited to share what we’ve been working on—a bot that automates the tedious task of processing unstructured documents from emails and entering them into ERPs. After many iterations, we’ve achieved 99.8% accuracy in extracting and mapping data from invoices, POs, and other documents. One surprising takeaway from this journey: building the AI was only 10% of the challenge! The real work came from handling edge cases, integrating seamlessly with various ERPs, and creating a reliable pipeline for real-world documents with messy formats. We’d love your feedback, thoughts, or questions about how we built this, the challenges we faced, or anything else. Let us know what you think! Thanks for checking it out! https://ift.tt/unMF4zD November 23, 2024 at 06:50AM
Show HN: Open-Source Pull Request AI Reviewer Hey HN, Over the last year, I’ve reviewed more than 1000 code changes. Most of the time was spent catching obvious mistakes rather than debating complex design decisions. If we estimate ~10 minutes per review, that’s 160+ hours spent reviewing code in just one year. So I thought: could I get some of that time back using LLMs? That's why I spent the last few weekends building Presubmit.ai, an open-source AI reviewer that runs as a Github Action right when you open a Pull Request. The results so far are promising: I estimate it can reduce the review time by 50%, which in my case would mean I save 80hours (~10 working days) per year. Unlike similar SaaS solutions, the goal is not to replace the human reviewer but to highlight obvious mistakes early, spot security vulnerabilities and give more context about the change. I like to think of it as a “pre-reviewer”. Some of its features are: * Line-by-line comments * PR summarization * Title generation on request * Responds to review comments It supports all major LLMs, but I’ve found Anthropic's Claude works best for this use case. Please give it a try and share your feedback! https://ift.tt/Vqm4lwo https://ift.tt/Vqm4lwo November 22, 2024 at 08:28PM
Show HN: Pull Request Reviewed by LLM This year I’ve reviewed more than 1000 code changes. Most of the time was spent catching obvious mistakes rather than debating complex design decisions. If we estimate ~10 minutes per review, that’s 160+ hours spent reviewing code in just one year. So I thought: could I get some of that time back using LLMs? That's why I spent the last few weekends building an LLM-based prereviewer that should take a first pass before the actual human reviewer. The results so far are promising: I estimate it can reduce the review time by 50%, which in my case would mean I save 80hours (~10 working days) per year. Linked above is an example of a PR where I'm testing the AI reviewer and it showcase how it can detect bugs, suggest best practices about token validity, generate summary and title, and even chat with me in review comments. The AI reviewer is a simple Github action that runs everytime I open or synchronize a pull request and you can see the source code at https://ift.tt/Vqm4lwo . https://ift.tt/sHuJWTE November 22, 2024 at 11:29PM
Show HN: VS Code extensions that display CGM blood glucose levels in status bar As a Type 1 diabetic, I need to continuously monitor my blood glucose levels. I’ve implemented a couple of Visual Studio Code extensions that retrieve the latest blood glucose readings from your CGM and display them in your VS Code status bar. One VS code extension uses the Nightscout CGM to retrieve the blood glucose readings. It requires users to run the Nightscout application on a hosted server. A nice benefit of Nightscout application is that it works with all the major CGM devices. However, a slight drawback of this option is that it requires a hosted third party software (Nightscout) for proper functionality. I’ve also implemented a Visual Studio code extension for those (like myself) that use the Freestyle Libre CGM. This version connects directly to LibreLinkUp to retrieve the latest blood glucose readings and display them in your VS code status bar. This removes the dependency for the intermediary Nightscout application. If you are or know any software engineers living with diabetes, these tools might be helpful with diabetes management. These are tools I built for myself that help me manage my diabetes. They are completely free and open source. I am not selling anything. Users of the tools can use them without any restrictions or connection to me. I am genuinely trying to help others in the community that are software engineers and might find this helpful. If you try out any of these extensions, I’d love to hear back from you. Any feedback on improvements are very welcome and appreciated. - https://ift.tt/E2kn6Q1... - https://ift.tt/dWwnsB5... November 22, 2024 at 12:19AM
Show HN: Autotab – Programmable AI browser for turning web tasks into APIs Hey HN, we're Alexi and Jonas the co-founders of Autotab ( https://autotab.com ). Autotab is a chrome-based browser you can teach to do complex tasks, with a simple API for running them from your app or backend. Here is a walkthrough of how it works: https://youtu.be/63co74JHy1k , and you can try it for free at https://autotab.com by downloading the app. Why a dedicated editor? The number one blocker we've found in building more flexible, agentic automations is performance quality BY FAR ( https://ift.tt/sh8x6y3... ). For all the talk of cost, latency, and safety, the fact is most people are still just struggling to get agents to work. The keys to solving reliability are better models, yes, but also intent specification. Even humans don't zero-shot these tasks from a prompt. They need to be shown how to perform them, and then refined with question-asking + feedback over time. It is also quite difficult to formulate complete requirements on the spot from memory. The editor makes it easy to build the specification up as you step through your workflow, while generating successful task trajectories for the model. This is the only way we've been able to get the reliability we need for production use cases. But why build a browser? Autotab started as a Chrome extension (with a Show HN post! https://ift.tt/Zqztwkm ). As we iterated with users, we realized that we needed to focus on creating the control surface for intent specification, and that being stuck in a chrome sidepanel wasn't going to work. We also knew that we needed a level of control for the model that we couldn't get without owning the browser. In Autotab, the browser becomes a canvas on which the user and the model are taking turns showing and explaining the task. Key features: 1. Self-healing automations that don't break when sites change 2. Dedicated authoring tool that builds memory for the model while defining steps for the automation 3. Control flows and deep configurability to keep automations on track, even when navigating complex reasoning tasks 4. Works with any website (no site-specific APIs needed) 5. Runs securely in the cloud or locally 6. Simple REST API + client libraries for Python, Node We'd love to get any early feedback from the HN community, ideas for where you'd like the product to go, or experiences in this space. We will be in the comments for the next few hours to respond! November 21, 2024 at 01:52AM
Show HN: Weave - actually measure engineering productivity Hey HN, We’re building Weave: an ML-powered tool to measure engineering output, that actually understands engineering output! Why? Here’s the thing: almost every eng leader already measures output - either openly or behind closed doors. But they rely on metrics like lines of code (correlation with effort: ~0.3), number of PRs, or story points (slightly better at ~0.35). These metrics are, frankly, terrible proxies for productivity. We’ve developed a custom model that analyzes code and its impact directly, with a far better 0.94 correlation. The result? A standardized engineering output metric that doesn’t reward vanity. Even better, you can benchmark your team’s output against peers while keeping everything private. Although this one metric is much better than anything else out there, of course it still doesn't tell the whole story. In the future, we’ll build more metrics that go deeper into things like code quality and technical leadership. And we'll build actionable suggestions on top of all of it to help teams improve and track progress. After testing with several startups, the feedback has been fantastic, so we’re opening it up today. Connect your GitHub and see what Weave can tell you: https://ift.tt/1JL7izw . I’ll be around all day to chat, answer questions, or take a beating. Fire away! https://ift.tt/1JL7izw November 20, 2024 at 11:13PM
Show HN: Browser-based website builder powered by LLMs I wanted to share what I've been working on - it's a AI site builder that runs in the browser powered by WebGPU and OnnxRuntime-Web. I have got the following all working to varying degrees: - text to code generation - image to code generation - microphone to text to code generation If you are on Mac for instance, it will interface directly with your GPU to power the LLM interface. It only requires downloading the models, and then everything after that is offline. It's not even close to as powerful as Claude or ChatGPT, but I like the idea of having the LLM run directly on your machine. I just did this for fun, but I am looking for a new role if anyone's hiring - https://ift.tt/Ec7DV1K ! More technical insight: - I also got the Typescript / React app to compile itself in the browser via a service worker https://ift.tt/7zgjQDr but took it offline due to some oddities with service workers. - A lot of the new speech models are a lot better than anything built into your phone / computer. I wonder when more computers will have them built in. - I added a CSP to the iframe only because I was worried about spamming sites since I update the iframe anytime a new token comes in. So if you have an image on the page it will get reloaded every time the iframe is updated. Otherwise there would be no reason for it. https://ift.tt/ACPqz78 November 20, 2024 at 01:54AM
Show HN: archgw: open-source, intelligent proxy for AI agents, built on Envoy Hi HN! This is Adil, Salman, Co and Shuguang and we're excited to introduce archgw [1], an open source intelligent proxy for agents built on Envoy [2]. Arch moves the critical but crufty work around safety, observability, and routing of prompts outside business logic. Arch is a uniquely intelligent infrastructure primitive, engineered with purpose-built fast LLMs [3] for tasks like intent detection over multi-turn, parameter identification and extraction, triggering single/multiple function calls, and offers convenience features to auto dispatch LLM calls for summarization based on data from your APIs via system prompts configured in archgw. Today, the approach to build a smart production-ready agent is weaving together a large set of mono-functional opinionated libraries, adding extra layers like LLM-based preprocessing to determine things like relevance and safety of the user's prompt (e.g. applying governance and guardrails). Once past that stage, developers must extract relevant information from the user prompt to determine intent, extract parameters as necessary, package relevant tools calls to an LLM to trigger a backend API to execute particular domain-specific task. etc. After all that is done then only are developers ready to trigger an LLM call for summarization and must manage upstream error handling and retry logic themselves. Not to mention, if they want to experiment with multiple LLMs or move between LLM versions, they have to write crufty undifferentiated code. This entire experience is slow, error prone, cumbersome, and not specifically unique. Prior to building archgw, the team spent time building Envoy [2] at Lyft, API Gateway at AWS, specialized search and intent models at Microsoft Research and worked on safety at Meta. archgw was born out of the belief that several rules based mono-functional tools should be converged into a multi-functional infrastructure primitive designed for prompts and agents. We built archgw on the highly popular, battle-tested open source proxy Envoy and re-imagined it for prompts and agents. For this we had to build blazing fast LLMs [3] that can handle crufty, ahead-in-the-request-path type of work in handling and processing prompts that are sent to an agent, so that developers can focus on what matters most: building fast personalized agents without the unnecessary prompt engineering and systems integration work needed to get there. Here are some additional details about the open source project. arghw is written in rust, and the request path has three main parts: * Listener subsystem which handles downstream (ingress) and upstream (egress) request processing. * Prompt handler subsystem. This is where archgw makes decisions on the safety of the incoming request via its prompt_guard primitive and identifies where to forward the conversation to via its prompt_target primitive. * Model serving subsystem is the interface that hosts all the lightweight LLMs engineered in archgw and offers a framework for things like hallucination detection of our these models We loved building this open source project, and our belief is that this infra primitive would help developers build faster, safer and more personalized agents without all the manual prompt engineering and systems integration work needed to get there. We hope to invite other developers to use and improve Arch. Please give it a shot and leave feedback here, or at our discord channel [4] Also here is a quick demo of the project in action [5]. You can check out our public docs here at [6]. Our models are also available here [7]. [1] https://ift.tt/7pHbr2N [2] https://ift.tt/Kv92WhE [3] https://ift.tt/xIYQwbK... [4] https://ift.tt/5BMEj3x... [5] https://www.youtube.com/watch?v=I4Lbhr-NNXk [6] https://ift.tt/zZiIJm1 [7] https://ift.tt/1S3Fi7x https://ift.tt/7pHbr2N November 20, 2024 at 12:56AM
Show HN: FastGraphRAG – Better RAG using good old PageRank Hey there HN! We’re Antonio, Luca, and Yuhang, and we’re excited to introduce Fast GraphRAG, an open-source RAG approach that leverages knowledge graphs and the 25 years old PageRank for better information retrieval and reasoning. Building a good RAG pipeline these days takes a lot of manual optimizations. Most engineers intuitively start from naive RAG: throw everything in a vector database and hope that semantic search is powerful enough. This can work for use cases where accuracy isn’t too important and hallucinations are tolerable, but it doesn’t work for more difficult queries that involve multi-hop reasoning or more advanced domain understanding. Also, it’s impossible to debug it. To address these limitations, many engineers find themselves adding extra layers like agent-based preprocessing, custom embeddings, reranking mechanisms, and hybrid search strategies. Much like the early days of machine learning when we manually crafted feature vectors to squeeze out marginal gains, building an effective RAG system often becomes an exercise in crafting engineering “hacks.” Earlier this year, Microsoft seeded the idea of using Knowledge Graphs for RAG and published GraphRAG - i.e. RAG with Knowledge Graphs. We believe that there is an incredible potential in this idea, but existing implementations are naive in the way they create and explore the graph. That’s why we developed Fast GraphRAG with a new algorithmic approach using good old PageRank. There are two main challenges when building a reliable RAG system: (1) Data Noise: Real-world data is often messy. Customer support tickets, chat logs, and other conversational data can include a lot of irrelevant information. If you push noisy data into a vector database, you’re likely to get noisy results. (2) Domain Specialization: For complex use cases, a RAG system must understand the domain-specific context. This requires creating representations that capture not just the words but the deeper relationships and structures within the data. Our solution builds on these insights by incorporating knowledge graphs into the RAG pipeline. Knowledge graphs store entities and their relationships, and can help structure data in a way that enables more accurate and context-aware information retrieval. 12 years ago Google announced the knowledge graph we all know about [1]. It was a pioneering move. Now we have LLMs, meaning that people can finally do RAG on their own data with tools that can be as powerful as Google’s original idea. Before we built this, Antonio was at Amazon, while Luca and Yuhang were finishing their PhDs at Oxford. We had been thinking about this problem for years and we always loved the parallel between pagerank and the human memory [2]. We believe that searching for memories is incredibly similar to searching the web. Here’s how it works: - Entity and Relationship Extraction: Fast GraphRAG uses LLMs to extract entities and their relationships from your data and stores them in a graph format [3]. - Query Processing: When you make a query, Fast GraphRAG starts by finding the most relevant entities using vector search, then runs a personalized PageRank algorithm to determine the most important “memories” or pieces of information related to the query [4]. - Incremental Updates: Unlike other graph-based RAG systems, Fast GraphRAG natively supports incremental data insertions. This means you can continuously add new data without reprocessing the entire graph. - Faster: These design choices make our algorithm faster and more affordable to run than other graph-based RAG systems because we eliminate the need for communities and clustering. Suppose you’re analyzing a book and want to focus on character interactions, locations, and significant events: from fast_graphrag import GraphRAG DOMAIN = "Analyze this story and identify the characters. Focus on how they interact with each other, the locations they explore, and their relationships." EXAMPLE_QUERIES = [ "What is the significance of Christmas Eve in A Christmas Carol?", "How does the setting of Victorian London contribute to the story's themes?", "Describe the chain of events that leads to Scrooge's transformation.", "How does Dickens use the different spirits (Past, Present, and Future) to guide Scrooge?", "Why does Dickens choose to divide the story into \"staves\" rather than chapters?" ] ENTITY_TYPES = ["Character", "Animal", "Place", "Object", "Activity", "Event"] grag = GraphRAG( working_dir="./book_example", domain=DOMAIN, example_queries="\n".join(EXAMPLE_QUERIES), entity_types=ENTITY_TYPES ) with open("./book.txt") as f: grag.insert(f.read()) print(grag.query("Who is Scrooge?").response) This code creates a domain-specific knowledge graph based on your data, example queries, and specified entity types. Then you can query it in plain English while it automatically handles all the data fetching, entity extractions, co-reference resolutions, memory elections, etc. When you add new data, locking and checkpointing is handled for you as well. This is the kind of infrastructure that GenAI apps need to handle large-scale real-world data. Our goal is to give you this infrastructure so that you can focus on what’s important: building great apps for your users without having to care about manually engineering a retrieval pipeline. In the managed service, we also have a suite of UI tools for you to explore and debug your knowledge graph. We have a free hosted solution with up to 100 monthly requests. When you’re ready to grow, we have paid plans that scale with you. And of course you can self host our open-source engine. Give us a spin today at https://circlemind.co and see our code at https://ift.tt/Nm9fdb7 We’d love feedback :) [1] https://ift.tt/Tq9SvV1... [2] Griffiths, T. L., Steyvers, M., & Firl, A. (2007). Google and the Mind: Predicting Fluency with PageRank. Psychological Science, 18(12), 1069–1076. https://ift.tt/XCvBITr [3] Similarly to Microsoft’s GraphRAG: https://ift.tt/ezTEvMl [4] Similarly to OSU’s HippoRAG: https://ift.tt/ql5mztj https://ift.tt/Zn3Bupl https://ift.tt/Nm9fdb7 November 18, 2024 at 11:13PM
Show HN: Store and render ASCII diagrams in Obsidian Obsidian plug-in that allows you to create and store ASCII diagrams in your notes. It can be used to visualise diagrams, flowcharts, complex tables, Gantt charts and more in technical documentation, that will be rendered as a nice SVG graphics. https://ift.tt/yN8PK1m November 12, 2024 at 07:33AM
Show HN: Next Beats – A Modern, Customizable, Open Source Lofi Music Player Hey everyone, I've always wanted to build my own lofi player, and I recently added it to one of my projects, You-TLDR. I decided to open-source it so more people can play around with it and make it their own. NextBeats is a modern, customizable lofi music player built with Next.js and TypeScript. It features a retro TV-style interface, YouTube integration for endless lofi streams, sound effect mixing (rain, cafe, etc.), and theme customization. Perfect for coding, studying, or relaxing, it lets you add custom channels, control music and effect volumes independently, and save settings locally. Open-source and mobile-friendly. Appreciate any feedback! https://ift.tt/PIDNsit November 17, 2024 at 11:01AM
Show HN: Windsurf – Agentic IDE At Codeium, we've been building AI-powered extensions for a while (we first launched our VSCode extension with autocomplete a little over 2 years ago!), but we've always thought there would come a day where we would hit the limits of what could be achieved within existing IDEs, so we decided to build our own: Windsurf (yes, it's yet another VSCode fork :) We've stuffed a lot of cool features into Windsurf—a super fast autocomplete model, an inline diff generation experience that feels truly native, but we're most proud of Cascade, which is an evolution of the sidebar chat experience that many other extensions have. Cascade can perform deep reasoning on your existing codebase, access a vast array of tools that allow it to run terminal commands and find relevant files, and it's omniscient of all the actions that the user has taken independent of invoking the AI. (You can for example, start implementing a change manually and just ask Cascade to "continue"). We've been using Cascade internally at Codeium on our actual production codebase, and we're getting actual value from it. We hope everyone here does too! You can find a bunch of demos of Cascade on our website but I want to show one that I made myself using Cascade to solve an interesting cryptography challenge: https://youtu.be/LbYepFmVB20 Cascade was able to explain the problem to me, install some libraries needed to interact with the challenge, give me some pointers towards a solution, and implement an attack that I described to it all by itself. https://ift.tt/HoXlORv November 13, 2024 at 10:39PM
Show HN: The Billy Bass I put on the Internet This one is just for fun. A couple years back I put together this tool to help me make synced up open/close mouth animations. End goal was always to "put it on the internet" like so many other things I remember when I was a bit younger. You can't drop this into a livestream these days without getting banned because people can't play nice, so I made it to record and distribute videos privately. I only code like this as a hobby, so it is a bit rough (I am a hardware guy), but these days you can get pretty far without knowing a lot. Enjoy https://ift.tt/nT7sU12 November 14, 2024 at 06:17PM
Show HN: Remy – AI-Curated Video Playlists on Any Topic Hey HN, we recently launched Remy, an AI agent designed to take the pain out of video search, and wanted to give the HN community a technical deep dive on how it works under the hood. The Problem: There’s a ton of valuable content on the internet, but finding the “best parts” of long videos is frustratingly inefficient. Current video search methods haven’t evolved much since the early days of YouTube and aren’t designed for today’s massive volume and variety of information. Instead, we’re left scrubbing through long videos or, worse, missing valuable content entirely due to decision paralysis. Remy’s goal is to offer a smarter way to surface exactly what you’re looking for, in a fraction of the time. The Solution: In less than a minute, Remy delivers custom playlists that isolate the best video moments from across the internet. It finds, clips, and organises segments to get you exactly what you need — without the endless search and skip game. How it works: Remy is powered by a stack of LLMs (and some non-LLM magic) designed for fast, focused video search and transcript processing. Here’s the pipeline: 1. Request Analysis When you send a message, the system decides whether to provide an immediate response, search the web, or start assembling video clips. If video is the best option, Remy generates a playlist outline with concise titles and detailed descriptions tailored to your query. For temporal queries, Remy automatically adjusts context to absolute dates (e.g., “tomorrow’s NBA games” → “November 15th NBA Games”). 2. Content Retrieval Using a ‘wide net’ approach, Remy generates a large set of targeted queries and searches the web for videos that could match your needs. 3. Multi-Step Filtration & Processing Each video goes through: - Transcript Pull: Extracts YouTube transcripts. - Non-LLM Filter: Filters out low-quality or AI-generated content based on YouTube stats, creator channels, release date, and other parameters. - Punctuation Restoration (BERT): Restores punctuation for better LLM comprehension. - Clipping: Uses an LLM to locate and clip the most relevant segments of the transcript based on your request. - Evaluation: Uses an LLM to score clips on relevance, completeness, and interest level. Only the best make it to the final playlist. 4. Reordering & Overview A final LLM gets fed the top 16 clips for each topic, filters them down to the best 4 (at most), and arranges them for maximum uniqueness. Each section gets a brief overview, with added context pulled and cited from the web to give you additional context. The result? A playlist tailored to your exact query, delivered in under a minute, without unnecessary noise. We’d love to hear your thoughts and feedback. If there’s something you’re curious to try or any edge cases that it's not handling, let us know. And if you come across any bad clips, please use the report button to flag them! Thanks https://ift.tt/yUe2qVn November 14, 2024 at 08:45PM
Show HN: Xfer, a data-transfer language Xfer is an experimental language I've been defining for a little while. It's intended to be a strictly-typed alternative to Json that offers a few other features, like comments, nested elements, placeholder substitution, and metadata. It's in VERY early days, so I'd love to hear your suggestions and feedback. https://ift.tt/ZwONhGW November 12, 2024 at 04:35PM
Show HN: Nango Auth – Free OAuth for 250 APIs Nango Auth is a free, source-available implementation of (O)Auth for 250+ APIs. Auth with external APIs is surprisingly difficult. OAuth is a mess[0], many APIs require parameters other than an API key or access token, and platforms like Stripe or GitHub even create their own custom auth protocols. We first launched Nango Auth on HN 1.5 years ago[1] when it only supported 40 OAuth APIs. Since then, it has expanded beyond OAuth into an end-to-end solution for API authentication: - Supports 250+ APIs[2] and almost a dozen auth modes: OAuth 2.0, OAuth 1.0, API key, Basic auth, and a half-dozen custom formats - Pre-built UI to help users pick & connect the API (including guides to find their API key and other required parameters) - Automatic credentials validation on connect - Secure, encrypted credentials storage & automatic token refresh - Detects expired/broken access tokens - Very detailed logs for quick & easy debugging More than 300 companies use Nango Auth in production for their product integrations: From fast-growing AI startups (Respell, Beam, Levity) to established SaaS players (Typeform, Semgrep, Electric) and even public companies. We also run a Slack Community[3] with 2,000+ engineers building product integrations. Nango Auth is a part of Nango[4], our source-available product integrations platform. While Nango itself is a paid product, our auth product (Nango Auth) is free for unlimited use, forever. It’s our small way of giving back to the community. You can either self-host Nango Auth or use our free cloud option. We hope Nango Auth can be helpful for your next integration, and look forward to your feedback! Demo video: https://ift.tt/h4f3zUq GitHub Repo: https://ift.tt/DurPSn2 Landing page: https://ift.tt/jMUstTF [0]: https://ift.tt/EDZ6lR7 [1]: https://ift.tt/62yocGn [2]: https://ift.tt/NZulbWc [3]: https://nango.dev/slack [4]: https://www.nango.dev https://ift.tt/jMUstTF November 13, 2024 at 10:21PM
Show HN: Foundation models for time series forecasting After months of brewing the perfect recipe in our AI kitchen, we're beyond excited to introduce Sulie - a fully managed (Model as a Service) platform for time series forecasting that actually works! From day one, we've had one mission: make powerful time series forecasting as easy as ordering your morning coffee. Now, businesses can make accurate forecasts from their data without the hassle of building complex models from scratch. We kept hearing the same frustrations from data teams trying to work with foundation models for time series forecasting: 1. "The zero-shot performance is about as reliable as a chocolate teapot!" 2. "Fine-tuning these models? Easier to teach a cat to bark!" 3. "And don't get me started on covariate support..." What makes Sulie special? • Automated model fine-tuning using LoRA - no PhD required! • Full covariate support for more accurate predictions. • Go from zero to production-ready forecasts in minutes (not weeks). • Zero ML complexity - we handle all MLOps heavy-lifting (you focus on the insights). We're already working with amazing customers who are: • Optimizing their supply chains • Making precise financial forecasts • Building custom models using Sulie's powerful embeddings And this is just the beginning! Stay tuned for deep dives into these use cases in the coming weeks Check us out: Python SDK: https://ift.tt/9uk0Ypo Website: https://sulie.co https://ift.tt/9uk0Ypo November 12, 2024 at 09:11PM
Show HN: Danswer APIs – Open-source APIs for building RAG apps over company docs Hi HN! I’m one of the creators of Danswer, the open-source enterprise search project that you may have seen on HN a bit under a year ago ( https://ift.tt/ZuWXiKf ). Really excited to be back to post about our new APIs :) Over the last half-year the #1 thing we’ve heard from our open-source community is the desire to build custom applications on top of the core connections with common apps + search + assistants. While this has always technically been possible, it hasn’t been easy for two main reasons: (1) inability to use APIs with auth-enabled w/o grabbing cookies from the browser and (2) a lack of simple APIs. Well no longer! For (1), we now provide a UI to generate API keys and for (2) we provide a pure “search” endpoint, a “chat” endpoint, and an implementation of OpenAI's Assistants API. With these, we take care of all the boring stuff (integrating with 35+ tools, constantly keeping docs up-to-date, and handling permissions) and provide an opinionated starting point for the generic hard stuff (really good hybrid-search + factoring in metadata, recency, etc.), so that you can focus on the specific business logic of the app. Back when I was at Alation, on-call for our team could have been made so much easier by a small, custom app that auto-responded to every alert based on Confluence, Slack, and a few Drive folders. Danswer’s APIs would power the core of this, and all I’d be left with are app-specific things like routing alerts to the right team member based on the document ownership of the cited documents. For an example of what it looks like to use the APIs, check out this script here: https://ift.tt/JDVprdq... ! If you're curious what's happening under the hood, check out our docs: https://ift.tt/V2KvwHu . Thanks, excited to hear your thoughts! https://ift.tt/ibxPyZm November 12, 2024 at 08:47PM
Show HN: Automatically tweet (X) the music you're enjoying on Spotify Hey HN! Excited to show a project called Bopping that makes it effortless to share the music your enjoying on X (Twitter). Bopping connects to both your Spotify and X (Twitter) accounts, tracks your listening activity, and automatically tweets out your top 3 tracks every day (timeframe adjustable in settings). When it knows an artists handle, it will tag that artist in the tweet, which has resulted in some nice artist/fan interactions. I often get artists retweeting or liking my tweets which feels very cool. It's been fun to see the tweets from the small group of people using it on my timeline. Would love for you to give it a try. Excited to hear feedback and see what tracks everyone’s vibing to. You can see some other bops here: https://ift.tt/ivcGwof... All bops are also added into a (rather eclectic) playlist: https://ift.tt/uglQt2m... https://bopping.to November 12, 2024 at 03:51AM
Show HN: Collaborative MIDI over WebRTC Hello Everyone Just finished developping a collaborative MIDI app with WebRTC The app unables you to send and receive MIDI signals over the net with peer to peer connection. You need to be signed up in order to use it and the repository is not yet accessible. https://ift.tt/MZPXYz9 November 12, 2024 at 02:29AM
Show HN: DocSend, but Made 10x Cheaper Hey HN, founder of Peony here. Most founders use DocSend to share investment or sales docs and have to pay $65 to $300 a month to get the security and analytics features. It pains me to see startups getting ripped off when the availability of technologies such as Clickhouse makes real time analytics almost a trivial feature to implement in 2024. We think it's time to end the carnage and democratize the technology. We set ourselves a high bar: we don't want to build another DocSend - we want it to be a far more superior product and also 10x cheaper. After a couple of months of work, I'm proud to share with you Peony - a first preview into what we believe will be the future of file sharing. In addition to the analytics and security features you'd expect coming from DocSend, we built Peony to be flexible and adaptable across all your file sharing needs and give you the ability to deeply personalize your file share per recipient. I'd love to hear what you think? We are the Prometheus stealing fire from the guarded Olympian mountains. Help us win the rebellion. https://peony.ink/ November 12, 2024 at 12:58AM
Show HN: Flash Kitty – Archive of Adobe/Macromedia Flash Movies from Flash Kit After realizing a few months ago that the current flashkit owners didn't really back up any of the user submitted movies, and getting some Flash nostalgia, I created this working Flash Kit archive using data from the Wayback Machine/Internet Archive. It uses Raffle so you can watch the submitted movies in a modern browser without needing plugins. It's not curated so you'll find a variety of things, some things are really creative and can be used for inspiration. https://ift.tt/ME3B92V November 11, 2024 at 11:11PM
Show HN: Krita RGBA Tech – Bringing Realistic Metal to Life in Open-Source Art Krita's v4.2+ RGBA brushtips let you paint with dimensionality, and it's a technology that Photoshop doesn't have (≖‿ゝ≖) It lets the creator do things like emulating light direction, making brushstrokes look 3D. I used them to make a set of metallic brushes. If there are any FOSS artists around, feel free to test them out. (p.s. Yes, I am learning Github so I probably made repository mistakes - apologies in advance lol) https://ift.tt/0C3j4eT November 11, 2024 at 08:15AM
Show HN: Open-source tool to convert ComfyUI workflows into web apps and APIs We built an open-source solution to transform ComfyUI workflows into production-ready web applications with minimal configuration. The project consists of: - ViewComfy (open source) - A tool for wrapping ComfyUI workflows in clean, user-friendly interfaces: https://ift.tt/FeRy5Wz - Cloud deployment layer - Includes an automatic dependency resolver that handles model management and node installation. Deploy either as interactive web apps (using ViewComfy) or REST APIs: https://ift.tt/Hs2wQvE Key features: - Zero-config deployment of any ComfyUI workflow - Automatic model/dependency management - REST API generation - UI builder components You can get started by dropping a workflow here: https://ift.tt/Hs2wQvE https://ift.tt/yH69O3G November 11, 2024 at 01:31AM
Show HN: I made a tiny device for automatically recording digital pianos Hey HN! A few years ago, I left my cushy big tech job to make hardware. And made the device I always wanted - an automatic piano recorder! I usually play piano improvisationally, and manually hitting record never meshed well with that. But there are always moments I wish I recorded, and now they are! Hopefully it scratches a similar itch for some of you as well! A few of the tech details: * built on an esp32-s3 * custom injection molded enclosure * BLE comms, sd card storage, DS1302 RTC * android & ios apps with Flutter * Shadertoy vfx support for video sharing - Chip https://jamcorder.com/ November 8, 2024 at 05:11AM
Show HN: I made a minimalistic AI calendar creator to accelerate daily planning Hey Hacker News! To test implementing AI (and paywalls tbh) in software for the first time, I made this simple calendar creator (no mail address required) that lets a person create a calendar based on his needs and availability and export it to his calendar app, pdf or receive a daily newsletter with his daily tasks (and more). After wrestling with different models (f.e. with huggingface) and encountering different problems, I decided to use ChatGPT's models, tune them a little and see how the result works out. ChatGPT's low price enables me to give users the opportunity to create calendars for free and without a mail address required. Demo video: https://youtu.be/UeZmqXcU5fY?feature=shared The website/tool includes: - Creating your calendar based on a title, date range and a description - Editing and "polishing" the calendar until getting a desired result - Exporting the calendar to your calendar app (can be Google Calendar, Apple Calendar, Microsoft Outlook, etc...) - Exporting the calendar to PDF - Signing up for a daily newsletter that contains a user's daily tasks, reminders, special ai notes and more! Some technical stuff: - I am using firestore database and paypal payments - I am always using the latest-fastest ChatGPT AI model (as generation-time is really important here) - The website is created with Nuxt 3 (Nuxt for vuejs is what Next is for reactjs) and designed with tailwindcss https://ift.tt/mCGAUJT November 10, 2024 at 10:42PM
Show HN: Dracan – Open-source, 1:1 proxy with simple filtering/validation config If you are tired of using Apache or Nginx to proxy your app and implement validation or limiting you may be interested in this hobbyist project. Thanks for all feedback. https://ift.tt/elcGjYR November 10, 2024 at 05:45AM
Show HN: Jaws – a JavaScript to WASM ahead-of-time compiler I've open sourced a JavaScript to WASM compiler. It's an experimental tool, but given the semantics I already implemented, I'm fairly certain I am able to eventually cover 100% of JavaScript spec. Any ideas, questions or critique welcomed! If you are interested in WASM, especially with new proposals like WASM GC or exception handling, it might be a good source of seeing these features in action - the project has a few thousand lines of hand written WAT so far. https://ift.tt/LhiY57N November 9, 2024 at 11:44PM
Show HN: RL Agent that can auto-optimize your LLM prompts Hey everyone! Along with my team, I've developed a reinforcement learning system that automatically optimizes LLM prompts, complete with a visualization feature to track both prompt structure and learning progress over time. Take a look here: https://ift.tt/sdt9biA... Check out our website too: https://ift.tt/sqJgQLh In terms of how this visualization works: The RL Prompt Optimizer employs a reinforcement learning framework to iteratively improve prompts used for language model evaluations. At each episode, the agent selects an action to modify the current prompt based on the state representation, which encodes features of the prompt. The agent receives rewards based on a multi-metric evaluation of the model's responses, encouraging the development of prompts that elicit high-quality answers. And see our github repo! https://ift.tt/4OKoFWb https://ift.tt/60WfRiA November 9, 2024 at 01:47AM
Show HN: Tuki – A GitOps tool for safely running one-off commands in production Tuki is a small, open-source utility that lets you run one-off commands in production through Git, bringing version control and code review to production scripts. I built it after accidentally dropping a production table due to a console typo Key features: - Run scripts through Git PRs instead of direct console access - Built-in auditing via Git history - RBAC through standard Git permissions - Simple setup for non-Kubernetes environments - MIT licensed Demo & more details: https://ift.tt/Mo0FgvT Announcement blog post: https://ift.tt/jhmHfpQ Would love feedback from the HN community! https://ift.tt/Mo0FgvT November 9, 2024 at 12:04AM
Show HN: ChatGPT for calorie tracking. Frictionless diet and exercise tracking Hi, I'm Steve and I've built this mobile app to solve my very long-standing problem of not consistently using any calorie tracking apps due to the poor UX (many clicks, too much friction). So, although I am not the best UI/UX designer, I have tried to remove friction at every corner with Journable. This applies to the 1. Tracking experience: Chat-based interface (think ChatGPT) to log food/exercise 2. Onboarding experience: No login screen, no measurements or goals, straight to the main chat window 3. Photo entries: Just take a picture of your plate and done That said, I'm not the best UI/UX designer so I would really appreciate any feedback on the app, and I welcome any and all comments & suggestions. Should I make it easier? Faster? Simpler? Does it need more color? Keen to hear it all! https://ift.tt/K7UkxXP November 7, 2024 at 11:52PM
Show HN: Intelligent Browser for Web Automation Powered by AI Hey HN this is Daniel, I'm excited to share Airtop with the community! Airtop enables AI agents to seamlessly browse and automate any website — even ones with complex authentication like MFA or OAuth. Whether you need to scrape, interact with dynamic content, or automate tasks across multiple browsers, Airtop's cloud-based infrastructure makes it scalable and accessible. Why Airtop? — AI-Driven Web Automation: Automate interactions with any site through natural language commands even if they include human-in-the-loop scenarios. — Authentication Ready: Airtop handles websites requiring authentication, allowing bots to perform complex tasks. — Scalability: Easily manage and deploy tasks, whether it's one browser or a thousand. — Compliance: Airtop ensures your data is handled with the highest security, availability, and privacy standards, giving you peace of mind while automating sensitive tasks. You can check it all out for free and play with our Studio (in the dashboard) to see how the API and prompts work. I would love to hear your thoughts and feedback! https://www.airtop.ai/ November 7, 2024 at 09:30PM
Show HN: Samsar One Version One is now live Hi all, 6 months ago I started building out an app which combines all the SoTA gen-ai models out there into one platform which anyone could use to create 2-4 min long videos starting from scratch, starting from an idea. I received a ton of feedback from HN community & others along the way. Today excited to announce version one of app.samsar.one, a platform which allows anyone to easily and consistently create long, coherent and consistent generative-ai videos while picking and choosing their model starting from merely an idea. https://app.samsar.one/ October 31, 2024 at 03:36PM
Show HN: NewsCatcher's Hyperlocal News API – Granular City-Level News Feeds We’ve launched NewsCatcher’s Hyperlocal News API to provide city-level news feeds for market analysis, localized apps, and data analytics. Why We Built It: Aggregating hyperlocal news is complex. Local news comes from diverse sources, many covering national topics, making it hard to isolate local events like transit updates, crime, extreme weather alerts, or infrastructure projects. Searching by location alone can lead to irrelevant results, especially with city names that overlap or match common words (e.g., Orange, Lake). What It Does: Our API provides precise, location-based news by processing 185K articles daily and tagging each with its associated_town down to the town level with 92% accuracy. Covering 31,000 U.S. locations, it uses advanced NLP to accurately associate location names, even when they’re common words. It enables you to: - Access near-real-time, location-focused news feeds - Filter by topic (e.g., crime, sports), keyword, and date - Cluster articles covering similar events and topics - Get sentiment score for a headline and article body - Recognize and extract entities Key Features: - AI-powered location recognition and validation - Advanced querying with boolean operators and proximity search - Multiple town association methods for precise location matching - NLP-enriched content for deeper analysis - Articles clustering - Multi-language support - High volume data retrieval (up to 1000 articles per request) For a demo, try our 15 public RSS feeds: https://ift.tt/UrcQuV6 . The API provides JSON output for easy integration. Tech Specs: - RESTful, JSON-based API - API key authentication - Scalable for high-volume queries We’d appreciate any feedback or questions on use cases. November 5, 2024 at 10:27PM
Show HN: Krixik – Easily sequence small/specialized AI models (pip-installable) Our initial intention was to experiment with a smartbook idea, but we couldn't find a good prototyping tool for small/specialized AI models. That rabbit hole took us through vector databases, through infra for small AI, and finally here. We’re particularly proud of the accessibility/simplicity of our code syntax. Krixik’s model library is limited to eleven model types (modules) and a few dozen models, but we will significantly expand it. Other enhancements, like a local client and several 3rd-party integrations, are also planned. You require API credentials to try Krixik. You can quickly get them through this form: https://ift.tt/QxDKWJL And here’s a demo video (our YouTube channel also has a couple dozen example videos in it): https://youtu.be/WpSSYLfvfdM https://ift.tt/4sOSxWQ November 4, 2024 at 08:33PM
Show HN: HN Push – Web push notifications for top stories on Hacker News I created HN Push to help reduce the urge to refresh Hacker News constantly. Since I rely on HN for real-time tech news, I wanted an efficient way to stay informed without having to check the site. Receiving summaries with Apple Intelligence was an added bonus. The source code is available on GitHub[^1]. [^1]: https://ift.tt/5GVw0I7 https://www.hnpush.com November 4, 2024 at 07:17PM
Show HN: P2P file transfer using WebRTC It's sometimes frustrating to send files among devices. Now you can send any file to any Chrome/FireFox/Safari client that wants to receive it: https://ift.tt/AUvi4GO Just share your ID, select a file and click send, and the receiving client will prompt to save it: There is no signup of any kind. You just need a way to copy the peer ID such as by email or text message, so they know who they're trying to get the file from. This makes it much easier to transfer files among devices. I tested it on Windows and iPhone, it worked perfectly. It does not use the server, it is peer to peer. https://ift.tt/ivPAyr4 November 2, 2024 at 08:19PM
Show HN: Autotab Instruct – Claude Computer Use with Guardrails for Reliability Hi HN, We’ve built a desktop app to create highly reliable AI agents that use a computer with mouse and keyboard. Until last week, we had tried many different approaches to open-ended agentic features but none of them had met our reliability bar. With Anthropic’s Computer Use this finally changed, and we just shipped a feature we’re calling Instruct. Instruct allows users to create agentic blocks as part of a larger Autotab skill that provides the structured logical flow to keep the automation on track. If you haven’t had a chance to try Computer Use yet, it is an impressive leap from the last generation of vision models (e.g. gpt4o struggles with relative positions, let alone coordinates). At the same time, it is still not good enough to be given a prompt and let loose. One of the big surprises to us early on was just how much intent specification is required for most real world workflows to run reliably. What looks at first like a simple form filling task usually turns out to have dozens of edge cases and super specific, hidden rules. Even human employees need to be shown how to perform these tasks, and then refined with question-asking + feedback over time. We wanted to build a tool for specifying intent, and iterating with the model to make it reliable enough for real work. - Automations run on top of an action scaffold, which works kind of like a very fuzzy programming language with strict types. This gives the model a high level plan that guides execution, and makes it easy to break out discrete steps to get the reliability you need. (Interestingly, this has also proven useful not just for the agent, but also for the human trying to create, verify and edit the automation.) - When the model is unsure it asks for clarification. For example, if you are in editing mode and the model thinks that an element looks meaningfully different than before, it will ask you to verify that it is the same element. - The agent has access to a memory system that lets it recall information from past runs as well as instructions and feedback from the user. Here's a short video of Autotab Instruct in action: https://ift.tt/cHJzZ0Y?... . There are a few more demos at https://twitter.com/autotabai/status/1852393973165199425 a75f06f82cab521bc78672ed35d85e8a. We’d love to hear what you think! November 1, 2024 at 10:26PM
Show HN: CKEditor 5 Ruby on Rails integration I've developed a new Rails integration for CKEditor 5 that embraces modern web standards. The previous ones are pretty dead right now. Looking for community feedback and improvements to make this gem even better - would love to hear your thoughts on essential features and optimizations. Let's keep Ruby on Rails alive :) https://ift.tt/6ToUNEg November 1, 2024 at 08:24PM
Show HN: Cerebellum – Open-Source Browser Control with Claude 3.5 Computer Use Hi HN! I was mesmerized by the Claude Computer Use reveal last week and was specifically impressed by how well it navigated websites. This motivated me to create Cerebellum, a library that lets an LLM take control of a browser. Here is a demo of Cerebellum in action, performing the goal “Find a USB C to C cable that is 10 feet long and add it to cart” on amazon.com: https://youtu.be/xaZbuaWtVkA?si=Tq9lE6BXv9wjZ-qC Currently, it uses Claude 3.5 Sonnet’s newly released computer use ability, but the ultimate goal is to crowdsource a high quality set of browser sessions to train an open source local model. Checkout the MIT licensed repo on github ( https://ift.tt/Bc4sFn5 ) or install the library from npm ( https://ift.tt/Vz6xS79 ) Looking for feedback from the HN community, especially on: What browser tasks would you use an LLM to complete? Thanks again for taking a look! https://ift.tt/Bc4sFn5 October 31, 2024 at 08:19PM
संदेश
