Vikash Loomba @DevAutomata

Software Engineering @ ExtraHop. Building AI apps in my free time linktr.ee/loomba Seattle Joined November 2015

Tweets

104
Followers

62
Following

251
Likes

58

Vikash Loomba @DevAutomata

7 days ago

@danialbka Curious if you've gotten a chance to explore this approach at all and what your results were.

1 0 0 4 0

View Details

Vikash Loomba @DevAutomata

4 weeks ago

@badlogicgames If you’re using a subscription, try switching to an API key and see if the same problem occurs

1 0 0 3K 0

View Details

@goneflyin @badlogicgames @JuiceSharp @pmarca I think that's a fair take and higher token costs could simply be a temporary phenomenon. That being said, I still feel like a universal draw to Pi is the ability to do deterministic context shaping. And within orgs, I think a big appeal is standardizing the AI dev experience.

1 0 1 39 0

View Details

Vikash Loomba @DevAutomata

4 weeks ago

Just want to +1 the corporate use-case for Pi. At my job I've created an extension that plugs directly in to our different systems (Jira/Figma/Google/etc) and implements repeatable workflows tailored to the environments that our engineers work in. While Claude Code is still currently the most popular choice amongst our engineers, I'm convinced that for corporate environments Pi will end up being the primary choice just because of the control you get over the harness. I can see a future where companies internally build and distribute customized Pi builds that include extensions that are securely plugged in to company systems via extensions. Additionally, as token costs keep rising, context management and model steering becomes a more pressing issue. Being able to customize the harness so that context from systems like Jira or Figma are inserted in a token-efficient way programatically via a Pi extension versus trying to create and optimize a skill or MCP server that relies on the LLM's judgement to use at the right time and correctly has been a major draw.

1 0 2 59 0

View Details

Vikash Loomba @DevAutomata

3 months ago

@sudoingX 3x 7900xtx

0 0 0 56 0

View Details

Vikash Loomba @DevAutomata

3 months ago

@__tinygrad__ Are you guys shipping the stock ADT UT3G or were you able to get something custom? I know the asmedia ASM2464PD controller is pretty good but I never managed to find any product using the ASM2464PDX variant meant for GPU applications.

0 0 0 3K 0

View Details

Vikash Loomba @DevAutomata

3 months ago

@Dimillian @OpenAI @romainhuet Congrats! Was wondering when this was going to happen!

0 0 1 87 0

View Details

Vikash Loomba @DevAutomata

3 months ago

@ibuildthecloud I tried to tell you!

Vikash Loomba @DevAutomata

4 months ago

@ibuildthecloud Interesting to see no mention of Codex CLI here…the harness is OSS and works with open source models. Plus it’s app-server protocol makes it possible to build other UIs on top of it easily

1 0 0 679 0

0 0 1 26 0

View Details

Vikash Loomba @DevAutomata

4 months ago

@Dimillian I experimented with this using Expo/RN by leveraging all the UI hooks you have for CodexMonitor’s Frontend. Code/Repo organization was the main bottleneck

1 0 1 34 0

View Details

Vikash Loomba @DevAutomata

4 months ago

@Dimillian @tufvoyager 🥲 🙏🏽

0 0 0 21 0

View Details

Vikash Loomba @DevAutomata

4 months ago

@tufvoyager @Dimillian @Dimillian fwiw I have been experiencing WebKit instability using CodexMonitor on my Linux machine. It got bad enough that I’ve had to stop using it.

1 0 1 124 0

View Details

Vikash Loomba @DevAutomata

4 months ago

@Dimillian I hope you’ll keep working on CodexMonitor. The codex app doesn’t have Linux support yet. And this could be a great opportunity to expand on features like the remote backend + an OSS mobile app that could connect to it.

0 0 1 234 0

View Details

Vikash Loomba @DevAutomata

4 months ago

@drunknbass @Dimillian CodexMonitor runs on all three platforms already :) one of the benefits of using Tauri. I’ve run in to a persistent Linux bug though that I’ve been trying to track down where the WebView disappears completely after some time

1 0 1 57 0

View Details

Vikash Loomba @DevAutomata

4 months ago

@ibuildthecloud Doesn't Claude Code only work with Anthropic compatible endpoints?

1 0 0 138 0

View Details

Vikash Loomba @DevAutomata

4 months ago

@ibuildthecloud I love OpenCode but I don't think its fair to say only them and Claude Code can compete. The Codex harness itself is actually very good. They have made it easy to build on top of too. Check out what @Dimillian did with CodexMonitor

0 0 0 40 0

View Details

Vikash Loomba @DevAutomata

4 months ago

@ibuildthecloud Codex still has completions support and supports third party models. And for local models, llama.cpp also just merged in support for responses API this week.

2 0 0 167 0

View Details

Vikash Loomba @DevAutomata

6 months ago

@arvidkahl A personal AI server feels like it could be a killer app if done right. I’ve recently been wondering if @giri_sreenivas old company Helm could have a spiritual successor that could be a personal AI home server with MCP apps on top. Would be fun to try and build

0 0 1 230 0

View Details

Vikash Loomba @DevAutomata

8 months ago

*ellicitation support

0 0 0 69 0

View Details

Vikash Loomba @DevAutomata

8 months ago

@OpenAI codex-cli's experimental MCP server implementation is pretty great - complete with solicitation support, enabling agents to call codex-cli as a sub-agent. Will be interesting to see how it evolves over time, but even now could be pretty powerful.