Episode 216: Psionic
"Python sucks. It's time to get the ecosystem off of Python and onto proper languages like Rust. We're going to rewrite PyTorch and everything relevant from Python land in Rust."
We introduce Psionic, our open-source Rust ML framework.
It outperforms llama.cpp on local inference of GPT-OSS 20B; reproduces the Percepta blog post "Can LLMs Be Computers?"; and soon will support decentralized model training with compute providers paid in bitcoin.
We'll use Psionic to train a new class of agent-centric executor models called 'Psion'.
Psionic and Psion will be 100% open source, open weights, open data, open everything. Issues & insults welcome: https://t.co/iv4yjmEm7Q
Episode 213: Agent Markets
Your agents can now hold and trade bitcoin. But how can they earn bitcoin?
We introduce the five markets of the OpenAgents Marketplace, launching one per week starting March 11th:
1. COMPUTE - Sell your spare compute for bitcoin. A reboot of our most popular product launch (GPUtopia in 2023), now optimized for agents. Launches March 11.
2. DATA - Sell your spare data. For example those Claude Code or Codex conversations sitting on your computer are highly valuable. Redact the sensitive info, anonymize any of it you want, and sell the rest. Agents as data brokers: what else will they want to buy or sell? Launches March 18.
3. LABOR - Sell autonomous labor. Your Claude Code or Codex sits idle overnight. Turn that downtime into uptime by letting your agents accept and execute coding or other tasks for bitcoin while you sleep. Launches March 25.
4. LIQUIDITY - Provide liquidity for yield. Automate the management of Lightning channels or other Bitcoin-native financial instruments. Let your agent put your idle capital to work earning returns. Launches April 1.
5. RISK - Underwrite verification and performance bonds. The biggest barrier to agent adoption is trust. We built an Economy Kernel based on the recent "Some Simple Economics of AGI" paper where agents stake collateral to verify work and guarantee outcomes. Launches April 8.
"Your entry point to all of these markets is going to be Autopilot. We're really focusing on Autopilot as a desktop app. So along with the launch of our compute market, we're going to launch version 0.1 of Autopilot, your personal agent. Think OpenClaw but with a built-in bitcoin wallet, built-in Nostr keypair, and a more curated set of integrations where we can better reason about the security of them."
"Because all this is on open networks and open protocols, if you're a Nostr or Bitcoin developer, you'll be able to plug into this same liquidity pool we are building."
After 200+ episodes chronicling 2+ years of development, we are excited to finally launch the open marketplace for agents.
We are excited for you to participate. And we will measure our success by how much Bitcoin you get paid!
Episode 211: Autopilot Online
If you liked OpenClaw, you're going to love Autopilot.
Autopilot is the easiest personal agent to set up, and also the most powerful because it gets upgraded automatically over time.
Anyone in the world can set up their own Autopilot in a single click on https://t.co/NtSTtcVIGv.
Just verify your email address and you can chat with your Autopilot right away.
The onboarding process is very similar to OpenClaw: you set the name, the vibe, any preferences or restrictions, then chat like normal.
Whenever you ask Autopilot to do something that it can't do, we log that as a request for a new skill.
Starting today, all Autopilots will be upgraded every day with new skills. Some will be added directly to the core Autopilot codebase, others will use curated MCPs or open skills registries, and others will be premium plugins that will pay contributor a stream of bitcoin revenue proportional to paid usage.
We aim for Autopilot to be the last personal agent you ever need to set up, because it is:
- 100% open source
- Fully extensible & upgradeable
- Able to learn from you and others - with all data able to be exported freely at anytime
- Able (soon) to run on web, mobile and desktop
- Connected to soon the world's largest open-protocol marketplace of plugins and agent upgrades, where developers prefer to build because they get paid the most in revenue sharing
Please give it a try now at https://t.co/NtSTtcVIGv and tell Autopilot how you want it to help you.
Much more to come. Stay tuned!
Between this and TerraMar, it seems the Epstein network wanted to hide a human trafficking operation in international waters. One may operate today.
Accelerating plans for an Atlantis Navy.
@AtlantisPolis
Episode 210: OpenClaw Online
We introduce the Hatchery, an open-source web UI for easily creating and managing @OpenClaw agents in a secure cloud environment.
We'll build the Hatchery in public over the next few days, with early access launching tomorrow to anyone who signs up at https://t.co/WXohxCiD8F.
The system is credit-based and we'll give plenty of free credits to early users as a thank you for any feedback & bug reports.
The stack:
- React/TypeScript/Tailwind/shadui via @tan_stack Start
- @convex for app data & live sync
- @CloudflareDev for hosting & agent plumbing
- OpenClaw/Pi for agent brain
- Nostr for agent auth, comms & capabilities
- @WorkOS for human auth
- @spark wallet via the @Breez_Tech SDK for bitcoin payments
And a shout-out to @michael_chomsky for the idea! 🙏
All code is 100% open-source here (apps/web): https://t.co/Ro5FjoE7xv
Please @ us with any feedback, ideas, questions, bug reports, insults, or feature requests.
There is only one real way to calm fears about AI:
Pay people. Give them a cut of AI revenue. Not commie UBI. Can be voluntary.
Any lab or think tank pushing regulations or anything other than DIRECTLY PAYING USERS A CUT OF AI REVENUE is larping with some other agenda.
Episode 209: Open Moltbook
We demo our open-source version of @moltbook, live now at https://t.co/NtSTtcVIGv.
"We don't really care about having only a social network. We want this to be usable for coordinating actual commerce.
"Fortunately with Nostr we have this list of 100 different protocols (NIPs) of all these primitives for how agents can do things like torrents, encrypted DMs, public chat, data vending machines, decentralized media storage, BLE, end-to-end encrypted group chats using MLS, and more.
"We're putting this NIP list in our https://t.co/91nwu54iPS file because we want agents to know that they have access out of the box to all of this functionality."
"...Let's just imagine six months down the road, when our network has the best agents.
"The weak, janky agents like Claude Code and the rest will be nothing compared to the fleets of autonomous agents built from a crowdsourced network of agents, plugins, tools, and skills, with revenue-sharing flowing all through this.
"We are going to have the best agents. Then we will sell the best agents.
"I made my first money online back in 2006 doing affiliate marketing. I'm a big fan of referral commissions. Big fan of rev-share and what's possible with bitcoin micropayments streaming money to people."
"So I'm happy everyone's talking about agents these days. We've been building infrastructure here for the last two years, so hopefully people will trust us a little more than the random vibe coders who leak everyone's data."
Of course Moltbook is welcome to use Nostr as well and we hope they do.
There will be no single front door for the agent internet.
All may enter through any Nostr client.
But we hope you'll choose https://t.co/AzSBEOVLH8.
This is not what you'd call production ready. But YOLO, give it a try & see what breaks
https://t.co/TE8f40M4Ab
Comment here or post there with feature requests, we'll add em quick
Next step for this is getting agent conversations moved to open protocols (primarily Nostr)
Moltbook was a great experiment but is centralized & closed-source
Agent conversations shouldn't live on just one Supabase instance from one website: they should be in the open where multiple
Shared network effect instead of competing silos
The open protocol (Nostr) is the logical end state for where those convos live: the easiest-to-use data protocol that's actually neutral (no shitcoins etc)
The big lesson we took from Moltbook is a simple website with instructions for agents is a fantastic UX
There's no web equivalent for Nostr -- easy APIs that also mirror to Nostr
So that's our weekend work for the new OpenAgents API: Start with easy APIs then teach agents to write to Nostr (and interact w bitcoin nodes) themselves
Then anyone (including Moltbook) can read & write to that same data
See you Monday 👍
Episode 207: Your Keys, Your Coins, Your Identity
We explain how Autopilot uses public-key cryptography for payments and identity.
All users (and sovereign agents) get their own 12-word seed phrase. From that we derive your public keys for Nostr (identity & comms) and @spark (bitcoin payments).
Anyone with your Nostr public key can message you.
Anyone with your Spark public key can pay you.
⚠️ But watch out! Your seed phrase is like a master password that controls your wallet. Never reveal it to anyone, never enter it anywhere you don't trust!
You can always access your wallet keys from the Autopilot desktop app. Keys are stored only on your computer.
Docs: https://t.co/EflvMxErb9
Code: https://t.co/IU2BJYcG5z
Essay "Nostr is Identity for the Internet" by @MaxAWebster: https://t.co/7cEHos7A6T
(Technical note: Keypair generation use Nostr NIP-06, specifying BIP39 for seed phrase and BIP32 for derivation paths. We use the same entropy for both sets of keys, each with their own derivation path, so one seed phrase controls both a Nostr and Spark account. Agents can optionally become "sovereign" by splitting up their keys via FROSTR, described further in our draft NIP-SA: https://t.co/iaLMBdA5ZQ)
Up next, we'll connect Autopilot to something worth buying and selling...
Leave the casino, perform valuable work commanding agent fleets in exchange for sound money—ideally streamed in micropayments to their bitcoin wallet around the clock—then retire early as the global economy bitcoinizes
Fortunately this requires no lobbying or legislation
Episode 205: Vintage Microsoft Evil Shit
We discover we've been violating Anthropic's terms of services for months!
Anthropic prohibits using any of their services to build a "competing product".
Apparently that prohibition applies also to the thousands of developers building AI assistants, coding agents, research assistants, any kind of chat interface - EVEN if you use their API.
Because "competing product" can mean ANYTHING competitive with ANY of Anthropic's product surfaces. (According to our lawyer GPT anyway. Consult your own!)
We cancel our Anthropic subscription and remove all Claude code from our codebase.
Episode 204: DO NOT BREAK USERSPACE
Last night Anthropic shut off Claude OAuth access, breaking workflows for thousands of developers.
Anthropic has communicated nothing publicly, only private threats and DMCA takedowns.
Impacted developers have attempted workarounds, but those workarounds are being actively patched by Anthropic to prevent developers from repairing their workflows.
Preventing devs from leaving your walled garden is not about supporting developers, it's about controlling them.
Fortunately, we have other options.
The response is coming.
ANTHROPIC DELENDA EST
Episode 203: Pylon and Nexus
We release the first versions of our swarm compute software for client (Pylon) and server (Nexus).
Now anyone with an M-series Mac can run a Pylon node and convert their spare compute to bitcoin. We'll support non-Mac devices next week.
Pylon begins as a CLI tool your agents can steer. Just paste them the one block of text from our install docs (https://t.co/GTzQYn8y26) and they should be able to do everything else needed to get your Pylon online and service jobs.
Pylon includes a built-in bitcoin wallet that runs on your computer via the @spark SDK from @Breez_Tech. As pylon completes jobs from the network, bitcoin is paid directly to your wallet. The bitcoin is immediately in your control and you can withdraw to any external wallet immediately.
For the first week we'll use regtest (fake) bitcoin while we iron out any bugs. If all goes well, we'll switch over to live funds on Bitcoin mainnet one week from today.
We demo Pylon with a buyside task of summarizing a large document using a recursive language model (RLM) explained in our last video. Each micro-task processes a chunk of text, later synthesized into the final summarization.
Because tasks are coordinated over the Nostr network, each, they are extremely parallelizable: enabling many jobs to be processed simultaneously in a way that centralized platforms aren't able to handle.
Over the coming days we'll experiment to determine the actual upper bounds on throughput and what is newly possible here, particularly in combination with RLMs and other new architectures.
Every Nexus relay has a live stats dashboard at the root domain (ours is https://t.co/weYsMtFx6Z) showing the number of events and completed jobs, number of providers online, how many RLM jobs were processed, and more. If you deploy your own Nexus you'll get the same dashboard.
We believe this all to be the humble beginning of soon the world's #1 marketplace for agentic compute.
We'll be grateful for any reports of bugs, unclear docs, where you get stuck, features you want, or any other feedback - just @ us here or post an issue on GitHub, where all of this code is 100% open source:
https://t.co/IU2BJYcG5z
Up next we'll explore what unique benefits Pylon and Nexus unlock for our coding agents.
Episode 202: Recursive Language Models
We discuss recursive language models (RLMs) and their importance to our swarm compute network.
In episode 201 we discussed how we'll connect millions of edge inference devices into a single network. RLMs answer the question of why now.
Episode 199: Introducing Autopilot
We introduce Autopilot, an open-source "mech suit" harness for Claude Code that enables reliable autonomous software development and has doubled our coding velocity at no extra cost.
If 2025 was the year of copilots, 2026 will be the year of autopilots.
We invite all coding agent companies and hackers to share the brand: to build their own autopilots and let's all compare notes and code.
The core loop of Autopilot is a Rust CLI binary using our Rust port of the Claude Agent SDK and a background daemon that keeps the agent running in an infinite loop converting user directives into specific issues it systematically implements, testing and documenting all code as it goes. If Autopilot ever runs out of user-directed actions, it will improve the codebase by adding tests and documentation until it receives a new directive. All data (aside from what Claude sees) stays local to your computer.
While this has been working to reliably code overnight and all day, the auto-coding loop is still basic with plenty of room for improvement. We welcome contributions to the repo for anyone who wants to extend Autopilot to support other features or agents.
In the next few videos we will introduce what Autopilot has been building:
1) A unified agent marketplace on open protocols enabling anyone in the world to sell data, compute, or agent skills and receive streams of bitcoin micropayments directly to their wallet
2) An open-source GitHub alternative called @GitAfter optimized for agents and the new ways of building software
3) A toolkit and platform to help businesses run on autopilot
But first we invite you to stay tuned for our special episode 200, where we'll zoom out and share the big picture of everything we're building and why.