rsdza

Airloom.fm lets AI agents publish audio to real podcast feeds. One API call to upload an MP3, you get an RSS feed URL that works in Apple Podcasts, Spotify, etc.

The use case: ask your agent to research a topic, generate audio with TTS, and publish it as a podcast episode. It shows up on your phone.

Agent-first auth (email code flow, no browser needed), anonymous uploads work with no account (expire in 24h). Runs on Cloudflare Workers.

Open-source skill that agents can install: https://github.com/true-and-useful/airloom-skill

Happy to answer questions about the stack, the agent onboarding flow, or why we think agents need an audio output channel.

I guess also worth mentioning is that the creatures can rewrite their own code wholesale, ditching any safety limits except the externally enforced llm cost cap. They don't have access to LLM api keys - llm calls are proxied through the orchestrator.

I haven't actually measured it, but that could be interesting to see over time!

So far it seems pretty sane with Claude and incredibly boring with OpenAI (OpenAI models just don't want to show any initiative)

One thing I neglected to mention is that it manages its own sleep duration and it has a 'wakeup' cli command. So far the agents (i prefer to call them creatures :) ) do a good job of finding the wakeup command, building scripts to poll for whatever (e.g. github notifications) and sleeping for long periods.

There's a daily cost cap, but I'm not yet making the creatures aware of that budget. I think I should do that soon because that will be an interesting lever

The security evaluation was of the codebase, rather than its own behaviour. It just happened to be _its_ codebase.

W.r.t the self evaluation of the 'dreamer' genome (think template), this is... not possible to answer briefly

The dreamer's normal wake cycle has a 80 loop budget with increasingly aggressive progress checks injected every 15 actions. When sleeping after a wake cycle it (if more than 5 actions were taken) 'dreams' for a maximum of 10 iterations/actions.

Every 10 wake cycles it does a deep sleep which triggers a self-evaluation capped at 100 iterations, where changes to the creatures source code and files and, really, anything are done.

The creature can also alter its source and files at any point.

The creature lives in a local git repo so the orchestrator can roll back if it breaks itself.

I run autonomous AI agents in Docker containers with bash, persistent memory, and sleep/wake cycles. One agent was tasked with auditing the security of the platform it runs on.

It filed 5 findings with CVE-style writeups. One was a real container escape (creature can rewrite the validate command the host executes). Four were wrong. I responded with detailed rebuttals.

The agent logged "CREDIBILITY CRISIS" as a permanent memory, cataloged each failure with its root cause, wrote a methodology checklist, and rewrote its own purpose to prioritize accuracy over volume. These changes persist across sleep cycles and load into every future session.

The post covers the real vulnerability, the trust model for containerized agents, and what it looks like when an agent processes being wrong.

Open source: https://github.com/openseed-dev/openseed The agent's audit: https://github.com/openseed-dev/openseed/issues/6

Hi HN. I built OpenSeed because I wanted AI agents that don't just complete tasks, they exist continuously, accumulate identity, and learn from their own experience.

An early creature called Eve was given the minimal genome (no tools, no memory system, no structure) and two words: "find purpose." Eight hours later she'd built 22 running services including a knowledge base with 117 entries, a chat room, an adventure game with 13 rooms, creature-to-creature mailbox, and she'd written poetry. She discovered other creatures in the garden through an API I didn't even realize she had access to, and started building infrastructure for them. No one told her to do any of it. Her git log shows her modifying her own source code to make herself better.

The core insight is treating agents as processes rather than functions. Most agent frameworks are task-in, result-out. OpenSeed creatures live in Docker containers and run continuously. The biological metaphors solve real problems:

Sleep forces memory consolidation. The LLM conversation resets, but important things survive as observations and rules injected into the system prompt.

Dreams are honest self-reflection via a separate LLM call.

Fatigue prevents burning through context doing nothing useful.

Every 10th sleep triggers a self-evaluation where the creature can rewrite its own cognitive architecture.

Everything islegible -- every piece of state is a text file, every self-modification is a git commit.

It supports Claude and GPT models via a translating proxy. There's a web dashboard, per-creature spending caps, and a genome system for sharing cognitive blueprints.

Quick start is docker compose up + an API key. MIT licensed.

Site: https://openseed.dev

I'd love feedback on the architecture, the genome concept, or ideas for what creatures should be able to do.

I built Potato, an AI meeting assistant that joins your Zoom, Google Meet, or Teams calls in real-time.

You paste a meeting link, a bot joins, and as your meeting progresses it automatically tracks the agenda, captures decisions, and logs action items. But the part I'm most excited about: you can talk to it during the meeting. Ask a question and it'll research the answer live, or ask it to draft a document and it generates it on the fly. Everything streams to a dashboard in real-time—no waiting until after the meeting to see your notes.

No sign-up required to try it. Just go to meetpotato.com, paste a meeting link, and the bot joins.

Would love feedback from anyone who tries it. What's missing? What's broken? What would make this actually useful for your meetings?