Building AI agent computers @AgentStationAI
Joined February 2015
- Tweets107
- Following583
- Followers204
- Likes202
Savannah Copland retweeted
We built our launch video in Claude Code using HyperFrames.
Now it's yours.
Open source, agent-native framework. HTML to MP4.
$ npx skills add heygen-com/hyperframes
RT + Comment "HyperFrames" to get the full source code of this launch video (must follow)
The best thing about building AI agent tooling is you can dogfood your product and automate your job at the same time :)
Planning to build a general purpose agent:
- I send an email with instructions to our agent email inbox
- Webhook launches an agent workstation via our API
- Agent assesses tools required (terminal, browser, etc)
- Agent executes task
- Agent sends email with results
Savannah Copland retweeted
Our AI Agent Workstations - complete with desktop, browser, and audio in/outputs - launch in milliseconds.
Here's how its done from our API.
You can monitor computer use + browser agents right from the @AgentStationAI UI.
Here's a demo of a simple browser workflow:
- Real-time logs
- In-browser VNC for monitoring / human-in-the-loop
- AI workstation hardware specs (we are currently running 8gb RAM / 4 vCPUs for default workstations)
Are folks using any AI tools for writing these days?
I'd love a cursor-like text editor that makes in-line edits, can help look for link references, save/manage files, and potentially refer to instructions / tone of voice examples / reference docs.
Given the Building Effective Agents article...
Are we finally aligned on the definition of AI agents?
- Agent: LLMs using tools based on environmental feedback in a loop
- Agentic Workflow: LLMs orchestrating / evaluating decisions in a pre-defined code path
Agree / disagree?
Finalizing more API endpoints today, and for whatever reason I forgot this means I can test them from the docs now!?
Why does that hit so much harder? 🤪
We use Docusaurus which has been mostly great. Search is up. The pieces are coming together! docs.agentstation.ai/api/int…
Savannah Copland retweeted
The uuidkey open source project got into the @golangweekly newsletter this week.
IF you care about API key design, and IF you appreciate beautiful things, and IF you like well-tested & performant Go packages... uuidkey may be right for you. 😁
The discourse around "AI agents are just RPA with LLMs" seems misguided to me.
Not only are agents already able to automate workflows that RPA can't, deploying RPA is time consuming. There's a lot of repetitive work that its not efficient to write and maintain scripts (or even Zaps) for.
If RPA is useful at the enterprise level to automate tasks at scale, AI agents will make it possible for SMBs and even individuals to outsource work.
Got our org Github page looking pretty.
Not sure why more folks don't use the public .github repo to make an org-level readme...
Copy my readme markdown/HTML if you like!
github.com/agentstation
In particular - these badges are a godsend - shields.io/
Added two quickstart demos for Puppeteer / Playwright to our Github.
You can use our AI workstations for fully autonomous "computer use" agents, but there's still a place for hardcoded actions. Use scripts for repeatable steps, and AI for everything else.
Link to our quickstart github project - github.com/agentstation/quic…
We published an open source project for "beautiful" API keys and wrote an article about it.
We knew it'd be controversial, as 1) beauty is subjective, 2) it sits at an intersection of topics developers know well, and 3) we wrote at length about a very small design decision.