Random Labs

We make coding agents built for autonomy.

[Random Labs @wearerandomlabs]

@dkta0 Not happening king 👑

[Random Labs @wearerandomlabs]

🤫

[akira @realmcore_]

Amazing.

Built in multiplexing is so good.

It’s like offloading a process from my brain.

The only issue is multiplexing in a chat is pretty annoying

[Random Labs @wearerandomlabs]

@dkta0 We've being building something very very cool

Also fixed all the bugs and added codex subscription support :)

[Random Labs @wearerandomlabs]

Opus 4.8 in Slate!

[Random Labs @wearerandomlabs]

We are actively working on making high touch interaction AND subagents independently more capable.

Stay tuned.

[Random Labs @wearerandomlabs]

To start off the week, we're breaking down brand new subagent features for subagent navigation:

The /subagents command allows you to explore all subagents

We default to letting the orchestrator manage the subagents but you can also interact with subagents directly!

[Random Labs @wearerandomlabs]

@JoshPurtell Feels good

[Random Labs @wearerandomlabs]

We've been hard at work improving the Slate CLI

Today we're releasing V.1.0.34 which includes:

- A fully rebuilt agent
- A brand new tui built in OpenTUI
- And much more

We're now able to support many of the features you've all been asking for!

[akira @realmcore_]

Byok in slate is finally happening btw

I know this was something you guys wanted since we launched in march but its a priority of ours now

We're building the most scalable agent around. 1000+ subagents runs easy in slate right now as it is. Multi day sessions are normal.

[Random Labs @wearerandomlabs]

FYI, OpenTui was built by the amazing team @anomalyco @thdxr

[Random Labs @wearerandomlabs]

@kagehiko Coming soon!

[Random Labs @wearerandomlabs]

@alhasaniq coming soon!

[Russell Joshua @russ98593]

@0xSero i would throw in @0xrandomlabs Slate as another one of those luxury line .

[Random Labs @wearerandomlabs]

@Orion_seeks It only gets more interesting from here

[Orion_seeks @Orion_seeks]

Threads and Episodes take us beyond RLM
@0xrandomlabs

https://t.co/wBcv2hFGBv

[Sauers @Sauers_]

The agent setup was originally Claude Opus running in a simple loop in Claude Code with some long prompt. I also tried Slate from @0xrandomlabs. It started out like MirrorCode: just implement Beagle (but in Rust). After a (low quality) initial implementation, I changed the goal to instead surpass Beagle and all other phasing and imputation tools on accuracy. I used a mix of Codex, Gemini, and Claude Code now with my direction on high-level goals (not fully autonomous). (They did not achieve the goal)

[Random Labs @wearerandomlabs]

@0xpolarzero @gajesh @FactoryAI True!

[polarzero @0xpolarzero]

@gajesh @FactoryAI https://t.co/ZVggFiuTw3 makes that super easy to code as a workflow.

@0xrandomlabs slate has this natively in their TUI.

[Random Labs @wearerandomlabs]

@emily00800800 System is designed to take advantage of model RL more by changing the internal context layout

[Random Labs @wearerandomlabs]

New release! 1.0.25 Contains:

Skills + skill-chaining
Agent fixes:
- Terminal use fixes
- Editing fixes
- Agent memory improvements
- Core harness leans into RL training more

[akira @realmcore_]

https://t.co/0xm0EmlygR

[Emily @emily00800800]

This twitter is transitioning to some kind of agent shitposting or idk but here goes