🗣️ Post 1: Show HN: Traceprompt – tamper-proof logs for every LLM call

As posted by: paulmbw  |  🔥 Points: 3

🔗 https://main.d1q2ygy4ts4vr5.amplifyapp.com

💬 Summary

Hi HN,

I'm building Traceprompt - an open-source SDK that seals every LLM call and exports write-once, read-many (WORM) logs auditors trust.

Here's an example - a LLM that powers a bank chatbot for loan approvals, or a medical triage app for diagnosing health issues. Regulators, namely HIPAA and the upcoming EU AI Act, missing or editable logs of AI interactions can trigger seven-figure fines.

So, here's what I built:

- TypeScript SDK that wraps any OpenAI, Anthropic, Gemini etc API call

- Envelope encryption + BYOK – prompt/response encrypted before it leaves your process; keys stay in your KMS (we currently support AWS KMS)

- hash-chain + public anchor – every 5 min we publish a Merkle root to GitHub -auditors can prove nothing was changed or deleted.

I'm looking for a couple design partners to try out the product before the launch of the open-source tool and the dashboard for generating evidence. If you're leveraging AI and concerned about the upcoming regulations, please get in touch by booking a 15-min slot with me (link in first comment) or just drop thoughts below.

Thanks!

🗣️ Post 2: Show HN: Zest – AI Agents for Slack

As posted by: kakaly040390  |  🔥 Points: 1

🔗 https://customer-d547k9rahyss1dfy.cloudflarestream.com/96768e95a601c45d2831a1532bfcdd66/iframe?preload=true&autoplay=true&poster=https%3A%2F%2Fcustomer-d547k9rahyss1dfy.cloudflarestream.com%2F96768e95a601c45d2831a1532bfcdd66%2Fthumbnails%2Fthumbnail.jpg%3Ftime%3D%26height%3D600

💬 Summary

Hi HN,

A few months ago, I started experimenting with building AI agents as Slack bots. What began as a fun weekend project has now evolved into a core part of how our team works every day. These bots live in Slack and handle real, non-trivial workflows such as:

- When a new signup comes in, one of them enriches the lead via Apollo, drafts a personalized email, and drops it into my mailbox.

- If someone mentions a new task in any channel, the bot jumps in, asks a few clarifying questions, and files a Linear ticket.

- Collecting customer info from Stripe for transaction failures and bringing it into the slack thread automatically.

- Scouring the web and drafting blog posts - sure you can do this with ChatGPT. But doing it collaboratively on Slack gives a different overall experience.

- We even made them proactive. For example, if someone talks about sprint planning, the bot jumps into the thread and asks if it should help break things down into tasks.

At first, this felt like "just automation." But what surprised us was how collaborative it became.

Since the agents live in public Slack threads, teammates started observing how others interacted with them — how people phrased prompts, tagged bots, corrected outputs. It turned into a sort of lightweight, social learning system. Everyone got better at prompting and thinking in “agentic workflows” just by being part of the conversation.

We also built a lightweight dashboard to help anyone on the team spin up and deploy new bots, manage access, customize behavior, and attach tools.

Demo: Here’s a short video of one of our bots in action — it self-corrects, navigates tools like Linear and internal APIs, and finishes the task end-to-end.

Stack:

- Vercel AI SDK

- axllm.dev(DSPy for typescript) for the agentic loop

- Composio for tool integrations

We are now running a closed beta for anyone interested in deploying Slack bots as agents. If interested, join our waitlist - https://heyzest.ai/

🎯 Final Takeaways

These discussions reveal how developers think about emerging AI trends, tool usage, and practical innovation. Take inspiration from these community insights to level up your own development or prompt workflows.