Replying to @JensenHuang
With love from Mercury
“The way to build an agent is to ask an agent to build it for you.” — Jensen Huang
So we stopped building Mercury's features and asked Mercury to build them.
Backed by Open source ❤️
1
15
Cosmic Stack retweeted
We built Mercury for People who want an Agent that does the Work, not one that just Talks about it.
nitter.cf/GithubProjects/status/…
We built Mercury for People who want an Agent that does the Work, not one that just Talks about it.
Imagine texting your computer a task from anywhere and coming back to it already done. This open-source AI agent lets you call, text or message it from the apps you already use, and it keeps working long after you've walked away.
2
4
627
Another model joins the Mercury lineup.
Baidu ERNIE 5.5 Flash is now available in Mercury Agent.
New model just pulled into Mercury.
Baidu ERNIE 5.5 Flash is now available.
So naturally, we took it for a quick test drive.
Explore Mercury → mercuryagent.sh
2
220
Replying to @HuggingModels @grok
It chose @mercury__agent over grokbot, there must be really something great about Mercury Agent.
3
2,101
Replying to @Lyra0x_
We’ve adopted DLQs on mercury bots, for regular chat this is always tricky.
Here we let the model decide to resume from DLQs if initial foundation was solid and has > 20% of work
In mercury code engine DLQs are always considered. Excluding planning
6
Worth a read if you’re building agents that need to keep working when nobody’s watching.
Mercury breaks down what can quietly go wrong overnight.
↓↓↓
Your AI agent worked perfectly while you were watching.
Then you left it running overnight.
A task gets skipped.
A run queues behind another.
The budget runs out.
Nobody tells you.
We went through Mercury’s scheduler and budget code to see what actually breaks when an agent runs unattended.
Read ↓
mercuryagent.sh/blog/running…
2
158
Replying to @alphakiloresrch @HuggingModels
Yes, the fact that its whole AI agent not just a notepad
18
Replying to @HuggingModels
Mercury Agent's second brain is not a typical second brain like other harnesses have, with thousands of markdown files and calling that a second brain. That's not a second brain at all.
At Mercury Agent, we have developed our own pattern- and prediction-based memory extraction, storing it in a structured way for faster retrieval: almost an O(1) retrieval rate through indexing and relational graphs.
1
4
144
Replying to @YasserAlotaiibi @mercury__agent
Imagine you want to remember when you last had pizza with your friend.
With many agents, finding that memory means going through hundreds of saved notes or files to find the answer.
Mercury Agent works differently. Think of it like a searchable memory. Instead of reading everything, it looks up exactly what you need and brings back the relevant information.
So rather than searching through an entire library, it's like asking a librarian who knows exactly where to look.
🤝
3
33






