发布 HN：Onyx（YC W24）——开源聊天用户界面

Hey HN, Chris and Yuhong here from Onyx (<a href="https:&#x2F;&#x2F;github.com&#x2F;onyx-dot-app&#x2F;onyx" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;onyx-dot-app&#x2F;onyx</a>). We’re building an open-source chat that works with any LLM (proprietary + open weight) <i>and</i> gives these LLMs the tools they need to be useful (RAG, web search, MCP, deep research, memory, etc.).<p>Demo: <a href="https:&#x2F;&#x2F;youtu.be&#x2F;2g4BxTZ9ztg" rel="nofollow">https:&#x2F;&#x2F;youtu.be&#x2F;2g4BxTZ9ztg</a><p>Two years ago, Yuhong and I had the same recurring problem. We were on growing teams and it was ridiculously difficult to find the right information across our docs, Slack, meeting notes, etc. Existing solutions required sending out our company&#x27;s data, lacked customization, and frankly didn&#x27;t work well. So, we started Danswer, an open-source enterprise search project built to be self-hosted and easily customized.<p>As the project grew, we started seeing an interesting trend—even though we were explicitly a search app, people wanted to use Danswer just to chat with LLMs. We’d hear, “the connectors, indexing, and search are great, but I’m going to start by connecting GPT-4o, Claude Sonnet 4, and Qwen to provide my team with a secure way to use them”.<p>Many users would add RAG, agents, and custom tools later, but much of the usage stayed ‘basic chat’. We thought: “why would people co-opt an enterprise search when other AI chat solutions exist?”<p>As we continued talking to users, we realized two key points:<p>(1) just giving a company secure access to an LLM with a great UI and simple tools is a huge part of the value add of AI<p>(2) providing this <i>well</i> is much harder than you might think and the bar is incredibly high<p>Consumer products like ChatGPT and Claude already provide a great experience—and chat with AI for work is something (ideally) everyone at the company uses 10+ times per day. People expect the same snappy, simple, and intuitive UX with a full feature set. Getting hundreds of small details right to take the experience from “this works” to “this feels magical” is not easy, and nothing else in the space has managed to do it.<p>So ~3 months ago we pivoted to Onyx, the open-source chat UI with:<p>- (truly) world class chat UX. Usable both by a fresh college grad who grew up with AI and an industry veteran who’s using AI tools for the first time.<p>- Support for all the common add-ons: RAG, connectors, web search, custom tools, MCP, assistants, deep research.<p>- RBAC, SSO, permission syncing, easy on-prem hosting to make it work for larger enterprises.<p>Through building features like deep research and code interpreter that work across model providers, we&#x27;ve learned a ton of non-obvious things about engineering LLMs that have been key to making Onyx work. I&#x27;d like to share two that were particularly interesting (happy to discuss more in the comments).<p>First, context management is one of the most difficult and important things to get right. We’ve found that LLMs really struggle to remember both system prompts and previous user messages in long conversations. Even simple instructions like “ignore sources of type X” in the system prompt are very often ignored. This is exacerbated by multiple tool calls, which can often feed in huge amounts of context. We solved this problem with a “Reminder” prompt—a short 1-3 sentence blurb injected at the end of the user message that describes the non-negotiables that the LLM must abide by. Empirically, LLMs attend most to the very end of the context window, so this placement gives the highest likelihood of adherence.<p>Second, we’ve needed to build an understanding of the “natural tendencies” of certain models when using tools, and build around them. For example, the GPT family of models are fine-tuned to use a python code interpreter that operates in a Jupyter notebook. Even if told explicitly, it refuses to add `print()` around the last line, since, in Jupyter, this last line is automatically written to stdout. Other models don’t have this strong preference, so we’ve had to design our model-agnostic code interpreter to also automatically `print()` the last bare line.<p>So far, we’ve had a Fortune 100 team fork Onyx and provide 10k+ employees access to every model within a single interface, and create thousands of use-case specific Assistants for every department, each using the best model for the job. We’ve seen teams operating in sensitive industries completely airgap Onyx w&#x2F; locally hosted LLMs to provide a copilot that wouldn’t have been possible otherwise.<p>If you’d like to try Onyx out, follow <a href="https:&#x2F;&#x2F;docs.onyx.app&#x2F;deployment&#x2F;getting_started&#x2F;quickstart">https:&#x2F;&#x2F;docs.onyx.app&#x2F;deployment&#x2F;getting_started&#x2F;quickstart</a> to get set up locally w&#x2F; Docker in &lt;15 minutes. For our Cloud: <a href="https:&#x2F;&#x2F;www.onyx.app&#x2F;">https:&#x2F;&#x2F;www.onyx.app&#x2F;</a>. If there’s anything you&#x27;d like to see to make it a no-brainer to replace your ChatGPT Enterprise&#x2F;Claude Enterprise subscription, we’d love to hear it!