Adib Mohsin

21, building - portable AI infrastructure for both enterprise and indie developers.

[Adib Mohsin @adib_builds]

Atlas lets you chat with your codebase using models that run locally on your machine or choose your favourite LLM provider

And see the complete decision tree of how your agent has progressed inside your codebase.

And yes it's completely free and open source!!

[Adib Mohsin @adib_builds]

Atlas packs all the tools to inspect how your coding agent's memory is progressing over time, now just imagine this at scale for teams where you can inspect each team members coding agent's history and messages.

A completely new way to blame, agent blame!

[Adib Mohsin @adib_builds]

Should I build a token gambling platform? Since no one can afford anything why should you be able to afford basic tokens let's just gamble it as well.

[Adib Mohsin @adib_builds]

Atlas - the open source alternative of @conductor_build and @notionhq has finally got support for multiple workspace.

Atlas was built around the need to document and observe how coding agents impact your codebase so you never lose track of what's going on.

And it's al OSS!!!

[Adib Mohsin @adib_builds]

Atlas 0.1.11 brings memory inspection for coding agents to see exactly how your codebase and git commits have been affected. How your coding agent manages policies and rules for your codebase etc etc

And the best part is you can query everything + this is completely open source!

[Emil Kowalski @emilkowalski]

An animaton doesn’t always have to involve some CSS transition or keyframes, it can also be as simple as swapping a character like we did for Linear’s /next page:

[Adib Mohsin @adib_builds]

I got tired of finding the perfect tool for vibecoding
my perfect workspace is a combination of @NotionHQ for documenting, @conductor_build for agent orchestration and embedded browser for researching (plus all the git tools I need)

And it's all open source!!!

[Adib Mohsin @adib_builds]

I turned my Design system into a Claude code skill
get started in two lines

/plugin marketplace add pacifio/ui
/plugin install atlas@atlas-ui

and get a dense UI for anything frontend.
It's a design guideline for agents, not a framework.

This is @shadcn for coding agents.

[Adib Mohsin @adib_builds]

My Agentic tool is really taking it's shape, heavily inspired from tools like @conductor_build and @athasdev

I am pushing a tool that allows me to research and plan first, contexts are the only differentiator in the era of coding agents.

[Adib Mohsin @adib_builds]

Everyone is building tools to manage more agent, I am building something to manage more context.

Browsing, Planning, Researching, Reviewing code, all from one platform.

Your personal Atlas.

Let me know how the UI looks, open sourcing on early-mid May

[Adib Mohsin @adib_builds]

Abstract, which derived from Cersei SDK in order to demonstrate Cersei's IO capabilities to write faster pipelines and invoke tool calling with minimal latency, clearly outperforms codex from @OpenAI in terms of generating code.

[Adib Mohsin @adib_builds]

@andrewjiang build your own agents at this point, that's what I am doing, jumping around and mixing multiple providers at once

[Adib Mohsin @adib_builds]

IO optimisations matter, while Cersei stands as the framework to build coding agents, the pipelines alone have made execution faster that's clearly noticeable, this demo shows how abstract (a demo tool for cersei SDK) dominates codex from OpenAI

[Adib Mohsin @adib_builds]

@OpenAI all open source
https://t.co/DXjrm4i35i

[Adib Mohsin @adib_builds]

IO optimisations matter, while Cersei stands as the framework to build coding agents, the pipelines alone have made execution faster that's clearly noticeable, this demo shows how abstract (a demo tool for cersei SDK) dominates codex from @OpenAI

[Adib Mohsin @adib_builds]

Everyone arguing about the Claude Code leak.

I turned it into a Rust SDK that turns the same architecture into an embeddable coding agent with graph memory on top. ~100us memory recall where Claude Code takes 7545ms, 2-4x faster for single prompt processing.

[Adib Mohsin @adib_builds]

@karpathy @__endif twitter as the message/knowldge/memory sync for agents, I am sure @elonmusk would invest in this

[Adib Mohsin @adib_builds]

@karpathy a very good feature to have for my coding agent framework, basically a complete knowledge dump in .md files anytime anywhere

https://t.co/DXjrm4i35i

[Adib Mohsin @adib_builds]

@karpathy @karpathy all open source of course
https://t.co/DXjrm4i35i

[Adib Mohsin @adib_builds]

@karpathy @karpathy I have been doing something similar, it's a research first tool, obsidian like knowledge base tool turned into a coding agent.

but I built a framework first that programmatically lets you control agents and it takes care of memory, MCP, skills, parallel agents.