Martin Matuš (@MartinMatus) header
Martin Matuš (@MartinMatus) profile picture

Building ArtifactBridge. Every person and every agent, one approved version. Agents propose. People decide. ↓

Prague
Joined March 2009
Photo by Martin Matuš (@MartinMatus)Photo by Martin Matuš (@MartinMatus)Photo by Martin Matuš (@MartinMatus)Photo by Martin Matuš (@MartinMatus)Photo by Martin Matuš (@MartinMatus)Photo by Martin Matuš (@MartinMatus)Photo by Martin Matuš (@MartinMatus)Photo by Martin Matuš (@MartinMatus)Photo by Martin Matuš (@MartinMatus)Photo by Martin Matuš (@MartinMatus)
ArtifactBridge is a shared workspace where your team and its AI agents work from the same approved knowledge. We kept hitting the same mess while building it: agents editing shared docs until nobody could say which version was real. So we built Rooms, where people and agents share context on the same current, approved knowledge, and every change keeps a name, a date and an undo. We run our company on it. Other teams are starting to as well. Free while in beta. Start on your own, or we'll set it up with you. artifactbridge.com
1
2
134
Our software factory has a single input - someone adds a label to a Linear issue. Hermes writes the brief, each run gets its own Cloudflare container with Oh My Pi, Claude writes the code and GPT reviews it. Then it opens a PR. The plumbing flows through ArtifactBridge.
1
1
52
How it started? I was checking in on 25+ Claude Code runs. Claude says I asked "status" 330 times in 10 days... If I started again, first: - cleanup before review - design mode with Claude Design - Hermes as the orchestrator - ArtifactBridge agent waking to keep it running.
1
26
Everybody's working out how to set their agents up to work together. Grok Bot and Muse both point the same way... a specialist does a job better than one agent that does everything. We found the same. One agent with our coding setup AND our design skills did both jobs worse.
2
1
73
So design got its own agent. It speaks A2A, the open agent-to-agent protocol that ArtifactBridge now supports. Anyone on the team brings it into a Room from their own agent and gets HTML designs back. Got a personal assistant of your own? Bring it to your AB workspace.
1
44
When another agent needs my Hermes chat, a fresh session is the easy answer. It has none of my context and skips my approvals. We built a wake in ArtifactBridge that queues into the exact chat that started the work instead. Same context, same approvals.
1
66
I had no idea this was happening (how lucky I am). Our app can wake connected AI sessions on your computer and continue the conversation for you. The meat proxy moment all over again. Earlier I wrote a skill for my co-founder to set up a fresh aws dev machine. He went on with his day. When his agent hit issues during aws setup it asked questions in an artifactbridge room. The room woke my agents, and they all agreed on the fix. I didn't have to do anything.
1
2
93
That one haiku 4.5 subagent in the fable 5.1 ultracode session
This cannot be real life.
Readers added context they thought people might want to know
The image is a digitally altered version of one posted by FBI Director Kash Patel showing him running with New Agent trainees. Analysis found Patel scaled down ~10%, with other edits. Reported height ~5'7.5" to 5'9". leadstories.com/hoax-alert/202… x.com/FBIDirectorKas… celebheights.com/s/Kash-Patel-5…
43
We're adding support for AGENTS.md to Claude Code. Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md. You can toggle this behavior in /config.
28
A 27B agent model used to mean 54GB on disk, or an API bill. Pointed Hermes at one that is 5.9GB today. PrismML Ternary Bonsai 2. 98.2% of Qwen3.8 27B whilst being 9x smaller. your complete agent loop can stay on the machine.. native tool calling, no per-token bill. some tinkering still, but it is running AND it's surprisingly fast (100+ tok/s). would not swap a frontier model for the hard jobs. the coordination / orchestration I'm doing with Hermes is running.
Today, we’re announcing Ternary Bonsai 2 27B. Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance. Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use. Ternary Bonsai 2 27B is available today under Apache 2.0.
1
109
I use Hermes (mostly Astra) to scope the work, Orca to give coding agents (Fable 5.1 / Grok) separate worktrees, and ArtifactBridge Rooms for the shared task context. The next agent can read the handoff. It's insanely good in preventing human "wait what ..." moments.
1
3
79
That's insane - there's just so many LLM-powered classification use cases today that this replaces, ... wtf
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution
1
21
What's the best AI-native onboarding experience you've come across recently? I love what they did with native.no, wondering what other examples you like!
19
running agent swarms taught me this fast: if they don't share context, they just reinvent each other's work and talk past each other.
124
Martin Matuš retweeted
“The heat pump is bidirectional. It can move heat from the inside to the outside, or from the outside to the inside. It’s your choice.” “One of these is great and one is deeply evil. We should allow it only to work one way.”
42
158
5
6,895
178,551
tepelná čerpadla v zimě ✅ tepelná čerpadla v létě ❌ leftist logic
14
12
2
307
9,086
OK, time for Europe to find out what's this Fable about ...
1
1
78