Anyone using claude cowork?

[Lss]

they can sign up enterprise plan. using thru api just for monitoring is kum lan

Abit more complicated but basically have another layer and we can switch models and pick what harness we like.

 

[testerjp]

I'm getting crazily good results from 5.6 Luna already. My spending is about 5 times lower than before.

I also realised max is counter productive for my work sometimes. Its like overthink until paralysis.

Also switching model ends up costing more also. Cheaper to just stick to 1.

Claude 5.5 is much better than 5 so i'm using it abit more.

But my credits for both Claude and Astra, both paid subs, used up 80 pct of weekly quota by Tuesday night.

For the latest update, i notice Astra a bit more reliable if need real world data.
Example Claude can fetch outdated stock prices.

But claude deeps to make slightly better deeper analysis.

For heavy agentic work, i think Deepseek flash V4.1 still okay,

One way i think is use Deepseek flash, or normal Chatgpt 5.6 sol do heavy work. let the frontier models check and explore.

 

[Lao_Tiko]

Up for use cases

You mean like setting a sandbox to do pen-testing with Kali Linux? That's probably my next project.

I haven't got the degree of confidence yet with my security stacks (so-sensor, ntopng, so-manager, wazuh-manager, crowdsec-lapi with a bunch of agents and bouncers, open-canary ... etc)

My Frigate with Hailo-8 accelerator, ollama wiring and OpenWeb UI is taking up most of my time now. Running nested dockers in LXC.

Yes, vibe-deploying since I have no IT experience.

 

[lalalalalala]

Abit more complicated but basically have another layer and we can switch models and pick what harness we like.

litellm ah? or openrouter. i guess if you want the flexibility then bo pian

 

[Lss]

litellm ah? or openrouter. i guess if you want the flexibility then bo pian

If tell you in-house you believe?

 

[Lao_Tiko]

litellm ah? or openrouter. i guess if you want the flexibility then bo pian

Using LiteLLM as the backend myself with OpenWeb UI so that I can use the embedded RAG. I don't have any paid tier subscriptions so I have a lot of downtime in-between. I usually take a break to drink beer in Geylang.

 

[Lao_Tiko]

If tell you in-house you believe?

RTX 3090 24GB?

 

[NintendoSwitch]

My company mandate through API. Coz they want to know what we are exactly doing bah.

 

[lalalalalala]

If tell you in-house you believe?

i believe, initially my company also do in house, until all the different devs want the different funny models, then we just ask them stfu and use openrouter / litellm

 

[testerjp]

RTX 3090 24GB?

Can meh, my pc happen to be 3090 24gb

i heard from AI like still not enough

 

[toolbox03]

You mean like setting a sandbox to do pen-testing with Kali Linux? That's probably my next project.

I haven't got the degree of confidence yet with my security stacks (so-sensor, ntopng, so-manager, wazuh-manager, crowdsec-lapi with a bunch of agents and bouncers, open-canary ... etc)

My Frigate with Hailo-8 accelerator, ollama wiring and OpenWeb UI is taking up most of my time now. Running nested dockers in LXC.

Yes, vibe-deploying since I have no IT experience.

No, like one man show business

 

[Lss]

RTX 3090 24GB?

Compute is in cloud but the software isn't standard bah

 

[Lss]

More of compliance bah. Super scared got attack or leak.

 

[Lao_Tiko]

Can meh, my pc happen to be 3090 24gb

i heard from AI like still not enough

That is actually very good liao.
You can run DeepSeek-R1-Distill-Qwen-32B and Qwen 3.5 32B locally
If your work requires flagship proprietary models then have to depend on cloud infra.

I run Qwen 2.5 on my 3060 12gb

Of course token generation speed is important lah. But typewriter 40wpm speeds is acceptable

I'd probably upgrade to your 3090 24gb when I find the money

Compute is in cloud but the software isn't standard bah

If you are running proprietary models like sonnet etc. no choice but go cloud. Other open source can run locally if you have the hardware

You can run API hooks on LiteLLM and choose either local or cloud models all through one OpenWeb UI interface

 

[Lao_Tiko]

No, like one man show business

One man system integrator+tech-support+deployment+coder+wrap mountain wrap sea qualify boh?

 

[Lss]

That is actually very good liao.
You can run DeepSeek-R1-Distill-Qwen-32B and Qwen 3.5 32B locally
If your work requires flagship proprietary models then have to depend on cloud infra.

I run Qwen 2.5 on my 3060 12gb

Of course token generation speed is important lah. But typewriter 40wpm speeds is acceptable

I'd probably upgrade to your 3090 24gb when I find the money


If you are running proprietary models like sonnet etc. no choice but go cloud. Other open source can run locally if you have the hardware

You can run API hooks on LiteLLM and choose either local or cloud models all through one OpenWeb UI interface

The arch I no say but can only say it feels designed to be adaptable. Even if tomorrow something new comes up we can almost immediately start using.

But I abit change until sian. Model aside UX already switch from vscode to cursor to Claude to chatgpt this year alone.

 

[toolbox03]

One man system integrator+tech-support+deployment+coder+wrap mountain wrap sea qualify boh?

So satki, chiu one man show NCS competitor?

 

[Lao_Tiko]

So satki, chiu one man show NCS competitor?

I only have one customer

But 24/7/365 response time

 

[testerjp]

That is actually very good liao.
You can run DeepSeek-R1-Distill-Qwen-32B and Qwen 3.5 32B locally
If your work requires flagship proprietary models then have to depend on cloud infra.

I run Qwen 2.5 on my 3060 12gb

Of course token generation speed is important lah. But typewriter 40wpm speeds is acceptable

I'd probably upgrade to your 3090 24gb when I find the money


If you are running proprietary models like sonnet etc. no choice but go cloud. Other open source can run locally if you have the hardware

You can run API hooks on LiteLLM and choose either local or cloud models all through one OpenWeb UI interface

how to run, your set up is what?

 

[toolbox03]

I only have one customer

But 24/7/365 response time

Jin satki, must be million dollar contract