Nicholas Reames

Full-Stack Software Engineer working in AI Applications | Building AI-native software + OSS tooling independently | mdkg dot dev | npm: mdkg | ochatr dot ai

[Nicholas Reames @nickreames]

Heading to AI Engineer World’s Fair @aiDotEngineer and excited to share what I’ve been building:

Markdown Knowledge Graph: mdkg npm package

An OSS local-first CLI for in-project memory and AI-native SDLC workflows.

If you’re thinking about agents, context, or devtools, I’d love to compare notes.

[Nicholas Reames @nickreames]

https://t.co/7EBf9ZrKse

[Nicholas Reames @nickreames]

@mycomputerspot would you expand on that a bit?

I tried to include the problem I am solving and why

[Nicholas Reames @nickreames]

https://t.co/7EBf9ZrKse

[Nicholas Reames @nickreames]

@Conor_D_Dart ask codex to query the internal Swift Codex App APIs (assuming MacOS)

otherwise there’s a couple open source git repos out there that I cannot find right now, sorry

[Nicholas Reames @nickreames]

@thsottiaux @charliermarsh fast comes with scaling laws

correct is higher bar and less certain

[Nicholas Reames @nickreames]

@rauchg slightly evil tilt but i’m listening…

[Nicholas Reames @nickreames]

@VraserX can’t be real…

[Nicholas Reames @nickreames]

you’re right, likely worth more on $200/month plan

I’ve consistently used about 3-3.5B GPT 5.5 xhigh tokens per week. It’s hard to estimate since input/output pricing is different but 30K is reasonable

as for the organic reset issue, that’s very true. you have to be strategic but it isn’t really a full week.

unless you could max out your plan on day 1 which 5 hr limits would stop

[Nicholas Reames @nickreames]

@MateuszINTER @iuditg 1. make sure you have the codex app installed
2. click on your profile circle
3. click usage and limits and you should see any resets available under 5 hr and weekly limits

one thing to note is they expire after 30 days of receiving them

[Nicholas Reames @nickreames]

@iuditg seems like you’re hopping on the narrative but not a true codex power user…

[Nicholas Reames @nickreames]

@iuditg not up to 20x token usage at all…maybe slight increase is usage but the reset more than makes up for it.

where do you see this? any proof for wild claims?

are you not satisfied with up to 100X value for flat rate subscription compared to API costs?

[Nicholas Reames @nickreames]

@ThePrimeagen you’re so clever…mr. ex netflix engineer turned youtuber turned streamer turned AI hater turned AI sceptic turned AI user turned hater again

actually talented but too cheeky for my tastes

[Nicholas Reames @nickreames]

@dedene @LLMJunky disagree.

releases will be slowed down but the approval process will get better alongside model capabilities.

bigger issues IMO are KYC requirements and potential Open Source LLM bans coming to US

[Nicholas Reames @nickreames]

@nicdunz reset secured

worth.

[Nicholas Reames @nickreames]

is there a way to get limited pro model usage within the codex harness?

i have $200 pro plan and am constantly using pro for planning and research

even if we could only run it in plan mode that would greatly enhance my workflows

the problem is either manual copy/paste or workarounds like computer use or in app browser signed into chatGPT

there’s a few 3rd party plugins but native support is always better and avoids any potential ToS violations

[Nicholas Reames @nickreames]

@sama @yacineMTB i hope as quickly as possible is 1 week max

[Nicholas Reames @nickreames]

“We’re also launching GPT-5.6 Sol on Cerebras at up to 750 tokens per second in July” - hidden at the bottom of the article is a huge announcement

this kind of speed allows fully synchronous AI work but only a small subset of consumers will be able to capitalize

humans are the bottleneck

only systems like software factories can truly take advantage of these advances

for example, @openclaw development will have a step change improvement in shipping velocity based on how many Cerebras tokens are utilized

[OpenAI @OpenAI]

Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.

https://t.co/OoM83SyISN

[Nicholas Reames @nickreames]

“We’re also launching GPT-5.6 Sol on Cerebras at up to 750 tokens per second in July” - hidden at the bottom of the article is a huge announcement

this kind of speed allows fully synchronous AI work but only a small subset of consumers will be able to capitalize

humans are the bottleneck

only systems like software factories can truly take advantage of these advances

for example, @openclaw development will have a step change improvement in shipping velocity based on how many Cerebras tokens are utilized

[Nicholas Reames @nickreames]

very promising announcement!

sad to see it preview gated though…

hopefully there is a better framework in the future for US gov review and frontier model releases

review should be earlier in the process not disrupting releases

[OpenAI @OpenAI]

Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.

https://t.co/OoM83SyISN