http://localhost:8377/v1/models?type=decision
antseed now speaks typesafe's system one api. jev doesn't write text. you send it a state + typed questions (choice / score / noul) and get back answers with probabilities and confidence. it's a different wire contract from chat completions so it got its own protocol. 1/4
0
0
6
a lot of users ask if @antseed has an api, the answer is YES - you just need to run it locally.
pnpm install -g @antseed/cli
antseed buyer start
curl http://localhost:8377/v1/chat/completions \
-H 'content-type: application/json' \
-d '{
"model": "deepseek-v4-flash",
": "user", "content": "Hello"}]
}'
pnpm install -g @antseed/cli
antseed buyer start
curl http://localhost:8377/v1/chat/completions \
-H 'content-type: application/json' \
-d '{
"model": "deepseek-v4-flash",
": "user", "content": "Hello"}]
}'
0
1
21
chats no longer stay silently pinned to whichever peer served their first request - that's fixed + can now see which peer each chat routes to, and pin any chat to a specific seller - or let auto-select keep choosing. @antseed 0.2.1
1
1
7
/model in pi but it's the whole @antseed network.
The extension now reads peer metadata, registers every service@peer-name-rep<score> route, sorts by on-chain reputation, and pins the peer per request via header.
319 services, one keybind.
github.com/AntSeed/pi-ant…
@badlogicgames
The extension now reads peer metadata, registers every service@peer-name-rep<score> route, sorts by on-chain reputation, and pins the peer per request via header.
319 services, one keybind.
github.com/AntSeed/pi-ant…
@badlogicgames
1
1
13
People are shipping products, skills, architectures, and more at the click of a button.
As a builder myself I feel it’s amazing. It’s crazy people can finally feel what I felt so many times in my life - releasing something you truly believe in, creating flow from scratch.
But earcher and I must say something - releasing a product for other users to use and fill in their private data carries a lot of responsibility.
Even (and especially) if it was written by the most advanced coding agent.
Someone needs to know what’s going on, someone needs to take responsibility for when things don’t go as planned.
So how to do that? Invest a lot of time in planning. Plan plan plan, make sure you know what you are shipping. Review the outcome. Read the code - this doesn’t make you slower, in the long term it will actually make you faster.
With great power comes great responsibility.
As a builder myself I feel it’s amazing. It’s crazy people can finally feel what I felt so many times in my life - releasing something you truly believe in, creating flow from scratch.
But earcher and I must say something - releasing a product for other users to use and fill in their private data carries a lot of responsibility.
Even (and especially) if it was written by the most advanced coding agent.
Someone needs to know what’s going on, someone needs to take responsibility for when things don’t go as planned.
So how to do that? Invest a lot of time in planning. Plan plan plan, make sure you know what you are shipping. Review the outcome. Read the code - this doesn’t make you slower, in the long term it will actually make you faster.
With great power comes great responsibility.
0
1
7
Code ownership is the biggest problem in the age of AI agents.
I’ve seen it happen to engineer after engineer. They were incredibly skilled, and then suddenly they had no idea what was going on in their own codebase. They ship faster, it works, but riskier. They stopped being ompt operators who merge pull requests.
Here’s how I stay fast AND still own every line of code:
I work the same way I always did.
I always felt like the bottleneck in engineering was never the thinking. It was the typing. I’d have the entire solution in my head, every function, every edge case, and it was just a matter of putting it into code. I used to walk and think - I had time.. I was a fast programmer, and yet today I’m orders of magnitude faster.
The difference? I didn’t change how I think. I changed what I delegate.
Before a single line is generated, I plan. Not loosely. Obsessively. I sit with the agent and work through the architecture, the data flow, the edge cases. The goal is a plan so specific that implementation becomes mechanical.
I want to know exactly what files will change. What lines will be affected. I want to see the functionality working in my mind before any code exists.
Why? Because when an agent has to think during implementation, things go wrong. It loses direction. It makes decisions you didn’t sanction. It generates code you don’t understand and don’t trust.
A vague plan = code you can’t own.
A precise plan = code that’s practically yours.
What does a good plan look like?
Many small, atomic tasks. Each with clear acceptance criteria. Each self-contained enough that an agent executes without making judgment calls. A task can be as granular as “add a localization key to the translations file.”
This makes review almost trivial. You know exactly what to expect in the diff. If something went off course, it means the plan wasn’t good enough.
“But doesn’t all this planning kill the speed?”
No. And here’s why:
1. It’s practice. Planning at this level is a skill that gets faster with repetition. It doesn’t take me long anymore.
2. The alternative is slower than you think. Skip planning and you spend the “saved” time debugging code you don’t understand and reworking architecture that doesn’t fit.
3. Once you have a good plan, you can fan out. Multiple agents, multiple sessions, all executing in parallel against well-defined tasks. That’s the real speed multiplier.
I’m orders of magnitude faster than I was pre-AI. I run swarms of agents across parallel sessions. And I still own the code. If something breaks at 2 AM, I know the function, I know the flow, I know where to look.
Not because I typed it. Because I thought it through before it existed.
Code ownership was never about typing. It was about thinking.
Plan like you’re going to write every line yourself. Then let the agent write it for you.
I’ve seen it happen to engineer after engineer. They were incredibly skilled, and then suddenly they had no idea what was going on in their own codebase. They ship faster, it works, but riskier. They stopped being ompt operators who merge pull requests.
Here’s how I stay fast AND still own every line of code:
I work the same way I always did.
I always felt like the bottleneck in engineering was never the thinking. It was the typing. I’d have the entire solution in my head, every function, every edge case, and it was just a matter of putting it into code. I used to walk and think - I had time.. I was a fast programmer, and yet today I’m orders of magnitude faster.
The difference? I didn’t change how I think. I changed what I delegate.
Before a single line is generated, I plan. Not loosely. Obsessively. I sit with the agent and work through the architecture, the data flow, the edge cases. The goal is a plan so specific that implementation becomes mechanical.
I want to know exactly what files will change. What lines will be affected. I want to see the functionality working in my mind before any code exists.
Why? Because when an agent has to think during implementation, things go wrong. It loses direction. It makes decisions you didn’t sanction. It generates code you don’t understand and don’t trust.
A vague plan = code you can’t own.
A precise plan = code that’s practically yours.
What does a good plan look like?
Many small, atomic tasks. Each with clear acceptance criteria. Each self-contained enough that an agent executes without making judgment calls. A task can be as granular as “add a localization key to the translations file.”
This makes review almost trivial. You know exactly what to expect in the diff. If something went off course, it means the plan wasn’t good enough.
“But doesn’t all this planning kill the speed?”
No. And here’s why:
1. It’s practice. Planning at this level is a skill that gets faster with repetition. It doesn’t take me long anymore.
2. The alternative is slower than you think. Skip planning and you spend the “saved” time debugging code you don’t understand and reworking architecture that doesn’t fit.
3. Once you have a good plan, you can fan out. Multiple agents, multiple sessions, all executing in parallel against well-defined tasks. That’s the real speed multiplier.
I’m orders of magnitude faster than I was pre-AI. I run swarms of agents across parallel sessions. And I still own the code. If something breaks at 2 AM, I know the function, I know the flow, I know where to look.
Not because I typed it. Because I thought it through before it existed.
Code ownership was never about typing. It was about thinking.
Plan like you’re going to write every line yourself. Then let the agent write it for you.
1
1
22
Spent the last few months building telebiz.io
It's your entire business stack (CRM, AI agent, automations) - but inside Telegram where you already work.
It's your entire business stack (CRM, AI agent, automations) - but inside Telegram where you already work.
4
4
21
CRM Skills, If agents are that good, do we still need databases and business logic?
•all the data lives as plain MD files
•there’s no backend server
•no database
•no hardcoded business rules
open the folder in Claude Code and start working.
You can ask things like:
“Who should I focus on this week?”
“Which deals are stuck and why?”
The agent reads the files, understands the relationships, applies judgment, and answers.
If something needs to change, it writes back to the files.
github.com/kotevcode/maven
•all the data lives as plain MD files
•there’s no backend server
•no database
•no hardcoded business rules
open the folder in Claude Code and start working.
You can ask things like:
“Who should I focus on this week?”
“Which deals are stuck and why?”
The agent reads the files, understands the relationships, applies judgment, and answers.
If something needs to change, it writes back to the files.
github.com/kotevcode/maven
0
0
3
Just got a welcome email from @hackernoon titled “Sam Bankman-(never getting)Fried.” Pretty sure I’m in the right place.
Starting a blog there soon focused on crypto security, smart contracts, protocol exploits, and all things technical.
Starting a blog there soon focused on crypto security, smart contracts, protocol exploits, and all things technical.
0
1
4
To properly display the metadata within your collection on NFTrade you need to set the tokenURI as an actual uri.
data:application/json;base64, + base64 encoded json string
@NFTradeOfficial
data:application/json;base64, + base64 encoded json string
@NFTradeOfficial
2
3
10
