AnythingLLM is a desktop application and self-hostable platform that lets you chat with your documents, run AI agents, and work with large language models entirely on your own machine. No account required. No data leaves your device unless you explicitly connect to a cloud provider. It runs on macOS, Windows, and Linux.

The core use case is document chat. You load in PDFs, Word files, CSVs, codebases, or content pulled from online sources, then ask questions against that material. The underlying LLM can be a local model running on your own hardware or a cloud provider like OpenAI, Azure, or AWS. You pick which one, and you can switch between them. It also supports multi-modal models, so working with images or audio is possible within the same interface.

Privacy is a first-class concern. By default, everything stays local: the model, the vector database, the embedder, your documents, and your chat history. Nothing is shared unless you configure an external provider. That makes it a practical choice for anyone handling sensitive business documents who can't send data to third-party APIs — including use cases that would otherwise call for legal AI platforms.

For teams, there's a self-hosted or cloud-hosted option with multi-user support. Each user gets isolated access, and admins get fine-grained controls over what users can see and do. The platform can also be white-labeled with custom branding, which makes it usable as an internal tool under a company's own identity.

Beyond document chat, it includes a built-in developer API for integrating its capabilities into existing products. If you're building on top of it, the API fits naturally into the broader LLM application framework ecosystem. A community hub provides shared agent skills, system prompts, and slash commands, so you can extend what the assistant can do without building from scratch.

The MIT license means the source is open and free to use or modify. If you want a lighter-weight alternative focused purely on chat, tools like LLMChat or LobeChat cover that narrower scope. AnythingLLM sits closer to a full local AI workspace, with document ingestion, agent support, and multi-user access all in one place.

Khoj is a personal AI platform built for people who want a capable assistant they can actually trust. It connects to your documents and the web, turning any supported AI model into a research partner, automation engine, and second brain. The focus is transparency: citations are verifiable, behavior is inspectable, and you stay in control of what it accesses.

At its core, Khoj is designed around three distinct use cases:

• AI second brain: Ingest your own documents, then ask questions, get summaries, or run research across both your files and live web sources. Agents can be built on top to handle recurring tasks.
• Research workbench: Read and organize academic papers with citations you can check. Useful for anyone tracking a field or doing structured literature review.
• AI co-worker (beta): Assign multi-step tasks, track their progress, and receive polished outputs. This runs locally on your machine, keeping sensitive work off third-party servers.

Because it's open source and self-hostable, you're not locked into a single model or vendor. Swap the underlying model, run it on your own infrastructure, and keep your data private. That's a real distinction from closed alternatives like Humanloop or proprietary chat tools.

Khoj sits in similar territory to LobeChat and LLMChat, but leans harder into autonomous agents and scheduled automations rather than pure conversational UI. The scheduled automation feature means Khoj can run tasks on a cadence without you initiating each one manually.

It's a good fit for researchers, knowledge workers, and developers who want an AI assistant with genuine depth rather than a thin chat wrapper.

Deta Surf is a browser and notebook rolled into one. It's built for people who spend serious time researching online and find themselves constantly juggling tabs, copy-pasting into notes, and losing track of where they found things. Instead of treating browsing and note-taking as separate activities, Surf makes them happen in the same place.

The core idea: your notes and your browsing exist in the same stream. You can pull in websites, PDFs, YouTube videos, and local files directly into a notebook and start asking questions or writing without ever leaving the page. Summaries, deep dives, and interactive graphs can be generated from the content you're looking at. No copy-pasting, no tab-switching.

Key capabilities include:

• AI summarization across YouTube videos, long PDFs, foreign-language content, and scientific diagrams
• Citations in notes that link back to the original source automatically
• Surflets, small interactive apps you can create to explain or visualize concepts without writing code
• Vertical tab sidebar to keep a large number of open tabs organized
• Local file import for PDFs, images, and documents via drag and drop
• Notebooks that collect websites, images, files, and notes together

Surf is designed to be local-first. Your data stays on your device, and you can import or export it freely. On the AI side, you can bring your own API key, connect to the LLM provider of your choice, or run local models entirely offline. That makes it meaningfully different from cloud-locked tools where your research and notes are held on someone else's server.

If you've used tools like Logseq or AnyType for networked note-taking, Surf occupies adjacent territory but adds a full browser layer on top. It's closer to a Comet Browser in spirit, but with a stronger emphasis on building a personal knowledge base from what you browse. The source code is open on GitHub, and the project is actively developed.