Humans use Chrome, Agents use Steel. curl -fsS https://xitter.cf/t.co/5wGwI3hXxj | sh
Joined August 2024
- Tweets631
- Following9
- Followers1.8K
- Likes1K
Steel retweeted
my hermes has been happily using @steeldotdev browser for months, but now finally everyone can easily add it as a plugin
Hermes from @NousResearch and @Teknium can now work through sites that are difficult to navigate.
The Steel browser plugin runs Hermes' tools in the cloud, with live viewing and page extraction.
It's Hermes browsing, turned all the way up.
See how to set it up ↓
Steel retweeted
Two commands:
hermes plugins install steel-dev/hermes-steel --enable
hermes config set browser.cloud_provider steel
Hermes asks for your API key. You get $30 in free usage credit on us. Add $10 in paid balance to verify your account for advanced features like proxies.
Steel retweeted
In just under 5 minutes, I configured a telegram bot that sends me apartment listings with @steeldotdev
Who wants a tutorial
Steel retweeted
harness for every purpose, and then harness application for every desire
with this sdk you can build your best research agent and feed it into your workflows
Launch week → Day 04
Today we're open-sourcing Atlas.
A deep research harness you can run, fork, and bend around your own domain.
🧵 h/t @junhssss
Steel retweeted
The "deep research" workflow bundled in Claude Code isn't deep. It's wide.
Niko pulled Claude Code's dynamic workflow to find exactly that shape. ↓
Steel retweeted
The team's been deep on the different shapes a browsing agent can take.
Agents fail in three ways on long runs: they quit early, grade their own work on a curve, and forget the rules after a compaction.
We'd been fighting them in a browser for months.
Three things we'd add from here 👇
Steel retweeted
Claude Opus 4.8 now ranks #1 on the OSWorld benchmark with an 83.4% score.
We just added it to the Steel leaderboard.
Congrats to the Anthropic team.
Introducing Claude Opus 4.8: it builds on Opus 4.7 with sharper judgment, more honesty about its own progress, and the ability to work independently for longer than its predecessors.
Available today at the same price.
ALT Benchmark table showing how Claude Opus 4.8 compares to its predecessor and to other models on tests of coding, agentic skills, reasoning, and practical knowledge work tasks.
Steel retweeted
This is what running browser agents in parallel looks like.
@nibzard watched Minority Report once and took it as a personal TODO.
spin up parallel Steel sessions, let the agents race, watch them all think on one screen
intervene mid-run, see them re-plan, get many answers back
@steeldotdev sessions are cheap and isolated, so the natural shape is many in parallel
minority report ui but for browser agents
Steel retweeted
We just need to give it a browser. Because, why not?
I had the wrong device. Older model, no speaker. Firmware wouldn't flash
so I pointed Codex at the repo and it fixed the firmware in one session
now just to wire in @steeldotdev and it could perform some web actions for me
the device is @M5Stack C Plus2: push-to-talk → @Cloudflare Worker → @Google Gemini
h/t @steveruizok for a fun project
Steel retweeted
Agents don't need a plugin ecosystem. They need CLIs.
Codex + @steeldotdev CLI + @resend CLI = personal newsletter
✦ Steel handles the browser complexity.
✦ Resend handles email sending.
✦ The agent handles the orchestration.
An agent that can browse the web AND send email is all you need.
Steel retweeted
An interesting use of our new Steel CLI 👇
> let an agent fight the messy web UI once,
> then export the exact steel browser commands as a deterministic bash runbook
> you and your agent can rerun it anytime
Steel retweeted
Verified template spotlight 🔦
Partnering with: @steeldotdev, the browser infrastructure for AI agents
Read more in their docs:
docs.steel.dev/overview/self…
Deploy with one click on Railway:
railway.com/deploy/steelbrow…
