Comments URL: https://news.ycombinator.com/item?id=49125532

Points: 33

# Comments: 2

You can connect it to Telegram.

The initial setup takes less than 10 minutes. All you need to do is:

1. Create a Telegram bot and get the bot token.

2. Sign up for Sanbox and connect your preferred model provider’s key.

3. Create a Sanbox API key.

4. Ask your coding agent, such as Codex or Claude Code, to set everything up for you using Sanbox CLI.

That’s it. Your Hermes Agent will run inside a Sanbox and be available through Telegram. You have full control over it. You can modify its configuration, add plugins, restrict network access, and more.

You can sign up here: https://sanbox.cloud

Comments URL: https://news.ycombinator.com/item?id=48951167

Points: 2

# Comments: 2

We are building Sanbox, a platform for running AI agents in isolated and resumable sandboxes.

We use the OpenCode SDK as the harness, support reusable templates, and have a CLI that works with Codex, Claude Code, Cursor, CI, or your terminal. Each sandbox has MicroVM isolation, a persistent filesystem, and live trail of run events. Can also self-host if required for security/compliance.

It's on the roadmap to add network ACL, secrets managements, LLM cost tracking & observability.

We are also happy to build custom integrations or triggers for specific use cases. For example, spinning up a sandbox from a database edge function when data changes, or connecting Sanbox to an email inbox for complex workflow processing.

It is still early, you can sign up to try here: https://sanbox.cloud

My email is also on my profile here.

Comments URL: https://news.ycombinator.com/item?id=48879908

Points: 5

# Comments: 0

Can you share more on what functionality you'd like from firewall? Feel free to drop me a message on oryx1729@protonmail.com if you'd like an early preview of my product.

The idea is to provide a great CLI experience so it's easy to use an AI client like Claude Code or Codex to launch sandboxes, fetch/retrieve data and prompts, review generated outputs, and manage queues for ask-user questions.

I also plan to add an AI gateway layer, similar to Tailscale Aperture(not affiliated but one of their engineers gave a great talk at AI Engineer it's on YT) for monitoring network access, distributing secrets, and controlling what services/systems a sandbox has access to.

For people building agent workflows, I'd love to hear feedback if you've tried other sandbox solutions.

What was missing and what was the hard part?

Comments URL: https://news.ycombinator.com/item?id=48835097

Points: 5

# Comments: 3

Comments URL: https://news.ycombinator.com/item?id=43822048

Points: 1

# Comments: 0

The idea is to let users ask natural-language questions like: - Are there new studies on treating acne? - What trials exist for migraine prevention? - Show me trials for managing hay fever symptoms - What are some ongoing trails for never-smokers with lung cancer?

If you’re curious, here are the links: - MedOryx App: https://med-oryx.streamlit.app - MedOryx GitHub: https://github.com/oryx1729/med-oryx

Would love any feedback or thoughts.

Comments URL: https://news.ycombinator.com/item?id=43756257

Points: 1

# Comments: 0

We built a live demo to compare DeepSeek R1 and OpenAI’s o1 model on a simple RAG pipeline using Haystack, our open-source AI framework. Both pipelines use the same retrieval and prompts to ensure a fair comparison.

Would love to hear your thoughts -- how do open-weight models stack up against proprietary ones for RAG?

blog post + demo link: https://www.deepset.ai/blog/deepseek-openai-comparison-by-ha...

Comments URL: https://news.ycombinator.com/item?id=42879717

Points: 1

# Comments: 0

GraphRAG transforms this experience with its entity-based query generation feature. By combining structured data (entities and relationships) with unstructured data (community reports and covariates), it crafts insightful follow-up queries tailored to a session's query history, highlighting critical themes and information.

With GraphRAG, users can be guided to uncover and explore insights from datasets.

Try out the live demo showcasing GraphRAG's potential in query generation: https://graphrag-demo.deepset.ai/query-gen

Comments URL: https://news.ycombinator.com/item?id=41102198

Points: 2

# Comments: 0

I've built a demo to query quarterly earnings call transcripts from a select few S&P 100 companies using Microsoft's new GraphRAG framework.

One of the key advantages of GraphRAG over traditional vector retrieval methods is its transparency and explainability. Unlike vectors, which are often obscure and not human-readable, the graphs generated by GraphRAG are easily interpretable.

In this demo, the graphs are used off-the-shelf without any domain-specific fine-tuning.

Check out the graphs here: https://bit.ly/46d757b Try out the demo here: https://bit.ly/3WfyvEX

Comments URL: https://news.ycombinator.com/item?id=41036519

Points: 2

# Comments: 0

Microsoft recently open-sourced the GraphRAG framework, which enables more contextual responses than traditional vector-based RAG, especially for summarization-focused queries on textual data.

However, a common critique is the LLM costs for constructing the knowledge graph. With the newly released GPT-4o mini, working with GraphRAG would now be ~30x cheaper.

We built a demo with quarterly earning call transcripts from a few S&P 100 companies comparing GraphRAG with GPT-4o, GraphRAG with GPT-4o mini, and Baseline RAG.

Try out the demo here: https://graphrag-demo.deepset.ai

Looking forward to your feedback!

Comments URL: https://news.ycombinator.com/item?id=41018636

Points: 5

# Comments: 0

Microsoft recently open-sourced the GraphRAG framework for information retrieval, utilizing graph-based structures. It automates the construction of knowledge graphs using LLMs and enhances retrieval by connecting related concepts and entities in a query for more contextual and accurate responses.

GraphRAG offers a larger connected context for retrieved information, which LLMs use to answer summarization-focused queries. It does not replace RAG but can significantly augment existing information extraction pipelines.

Asking questions on financial data is one example of a great use case for GraphRAG.

Check out this demo comparing quarterly earnings call transcripts from a few companies to see a side-by-side comparison: https://graphrag-demo.deepset.ai

Curious to hear your thoughts.

Comments URL: https://news.ycombinator.com/item?id=41006815

Points: 3

# Comments: 1