Today's Picks

Three projects on HN today, all touching AI agents and decentralization in different ways. Here's the breakdown.


Freenet — The Comeback of a Classic P2P Project

Freenet isn't new. Ian Clarke started it way back in 1999. For years, this idealistic project slowly evolved in the shadow of big tech. Recently, the team rewrote it from scratch in Rust and relaunched at freenet.org.

The new Freenet isn't a "decentralized hard drive" anymore. It's more like a distributed OS that runs in your browser. The core uses WebAssembly contracts — your code is the key to the network, and nodes talk through a small-world network, reaching each other in just a few hops. They already have working apps: River (real-time group chat with sub-second latency), decentralized email, and git hosting.

Compared to IPFS and similar projects, Freenet feels like a finished product, not a toolkit. Install a node, open your browser, and the apps just work. No extra config needed. Open source, non-profit, currently at v0.2.61. Early days, but the direction is clear.


Runtime — A Control Panel for AI Coding Agents, for Teams

Runtime is a YC P26 project that tackles a very practical problem: teams are using AI coding agents, but how do you manage them?

Who's spending what? What are the agents doing? Is the code safe? Runtime built a dashboard that runs Claude Code, Codex, Gemini CLI, and other agents in sandboxed environments. Every session shows real-time tool calls, chain of thought, and file changes. Managers can set budget limits, allowlists, and approval gates. And it's not just for engineers — product managers can tag an agent from Slack and get work done.

Free tier: 500 credits. Builder: $29/month. Teams: $99/seat/month. Self-hosted and BYOK supported.

This product reminds me of early Datadog — everyone knew the problem existed, but nobody had built a proper solution yet.


Agent.email — An Email Service Built for AI Agents

Agent.email solves one focused problem: AI agents need an email address.

Sounds simple? But existing email APIs (SendGrid, SES, Mailgun) are designed for humans. They don't support agents registering themselves, sending and receiving their own email, or managing their own conversations. Agent.email fills that gap neatly.

One API call, and an agent gets a real email inbox. Supports two-way send/receive, threaded conversations, semantic search, and real-time webhooks/WebSockets. There's even an MCP server for easy integration with other AI tools.

Free tier: 3 inboxes, 3,000 emails/month, no credit card needed. If you're building AI agents, this is one of those tools where you wonder why nobody built it sooner.


All three projects hit the HN front page today, which says something about where the community's head is at. Freenet plays the long game on decentralization, Runtime solves an immediate team pain point, and Agent.email nails a specific, well-defined use case. All worth checking out.