CJS

"claude usage limit reached. your limit will reset at 7pm"

Ah makes sense, seen similar when it tries to use my custom explore droid.. so I guess before starting a mission just remove any BYOK custom droids you have to avoid it, I wonder if setting a hard constraint during the brief of “only use worker droid and validator no custom sub agents” would work

@cjayls @tomosman @matanSF @FactoryAI @EnoReyes Actually, missions is somewhat different; so when you’re in a mission, BYOK models work for the worker, the orchestrate and the validator… if any of those agents try to spin up sub agents, the sub agents fail due to BYOK failing with sub agents.

Shifted one of my projects over to @FactoryAI this morning and ran two missions side by side. We're 6hrs and 5hrs respectively with minimal input from me apart from confirming the plan and adding a couple of keys a few hours in. Stunning work @EnoReyes @matanSF + team! (and you @bentossell 😅)

@mikez93 @tomosman @matanSF @FactoryAI @EnoReyes Oh they can only be used for the orchestrator??

@tomosman @matanSF @FactoryAI @EnoReyes It really is amazing… it would be even more amazing if BYOK models could be used in sub agents👀

@nico_jeannen @AnthropicAI First time hitting it myself and in one day lol.. is opus 4.6 1m context window now just that more expensive towards usage? Gonna have to go back to literally only using it when I need frontend tweaks and just using codex for everything else

Idk what's going on with @AnthropicAI but they messed up the weekly limits badly it seems Anyone else with the issue? I've almost never hit the weekly limit on the x200 plan (except maybe 2-3x a few hours before the reset) and now I'm hitting it on Monday lol

@rudrank @Dimillian Now just wire in openai imagegen in the background to help us create nice App Store screenshots EZ

Yet another App Store Connect client

I would go out of my way to not use a service if it’s website looked like this lol

This Dude is the next Dave Chapelle. It take a special person to come near the line, stomp on the line, cross the line, and still not offend nor be offensive 🤣😂🤣

@TheHiddenOneAC How does the shield suddenly appear though? Just out of thin air into your hand? I think that’d actually make it less immersive for me tbh but need to try it out

The 2 Quality of Life mods 'Remove Shield' & 'Closer Camera' elevates Crimson Desert for me quite a lot. Makes me feel more immersed in the game.

@idonotexistelol @scary900 Is it public info how much black desert cost to make / maintain & the revenue from it VS crimson desert so far? Wonder if they’ll prioritise more single player games in future

@scary900 Them being live service Devs makes so much sense

Just read the patch notes, they're literally patching stuff I saw people complaining about not even 24 hours ago

@chetaslua Would be good if they released what % of pro users on average this will impact based on their usage

🚨 Chatgpt 100$ plan is coming soon One bad news - now the 200$ plan is not truly unlimited ( it's 20 times the plus uses) Good news - at least more people can use early features at a cheaper rate

@HarryStuck77 @droid What model choice for orchestrator and worker then? Does validator push back against poor code choices/design even if it achieves what the validation contract requires?

The Validator in @droid In Missions, the validator is the checkpoint worker, not the builder. Feature workers do the implementation for features, while validator workers run at the end of each milestone to verify that milestone’s work. Factory’s docs say the milestones define validation frequency. For simple projects, one milestone can be enough; for longer or more complex projects, more frequent milestone validation helps keep the foundation stable. What it does in practice Factory does not publish a full internal validator checklist on the Missions page, but they do say validation workers verify the work, and that validation can surface issues that cause the orchestrator to create additional fix features. So the useful mental model is: the worker says, “I built it,” and the validator asks, “Did this milestone actually land, or do we need correction before moving on?” Why it matters The validator matters most when the project is large enough that bad work compounds. Factory’s guidance for large features emphasizes incremental validation and testing at each phase boundary instead of waiting until the end, because phase-by-phase validation reduces drift and expensive rework later. That is exactly the validator’s job inside Missions. Cost and speed impact Validation is not free. Factory’s Missions docs say to budget roughly one feature-worker run per feature and one validator-worker run per milestone, and they give a rough heuristic of total runs ≈ #features + 2 * #milestones. They also note that this is only a floor, because validation can uncover problems that trigger more follow-up work. Why you should care about model choice If the worker is the one that makes changes, the validator is the one that protects you from quietly wrong progress. That means validator quality matters more than people think. Factory’s current model guide ranks Claude Opus 4.6highest for depth and safety, Claude Sonnet 4.6 as a strong balanced daily driver, and GPT-5.4 as excellent for large-context tasks; by contrast, cheaper Droid Core options like GLM-5 and Kimi K2.5 are None-only for reasoning, while MiniMax M2.5 is the strongest cheap Droid Core option because it supports Low/Medium/High reasoning. Based on that guidance, I would use a stronger reasoning model for validation than for cheap bulk work.

@EnoReyes Thanks helpful as always!

The readiness report is more of a goalpost. You can def run a mission without it, but the mission is tuned to seek the types of validation the readiness report validates for you. It’s all tied together in a way that when a motivated organization combines turns into differentiated value from agents.

Couldn't resist asking droid to make this chart - our team of 25 technical staff is moving quite fast and shipping every day. This doesn't even include the bugfixes, reliability, tests, research, evals, internal apps, etc. that we're working on!

@richychn @FireworksAI_HQ @mweinbach @lumendriada Thanks for quick response! Legend

@cjayls @FireworksAI_HQ @mweinbach @lumendriada you can click manage and then toggle auto-renewal. sorry! we'll be making the experience smoother. priority was to protect people from unexpected charges

@mweinbach @lumendriada Open sign up for now, but we have limited capacity. We have a little progress bar on the site that shows you availability.

Host: Sir, do you know if Iranians are starving? Trump: Yeah I do. But you’re so sexy.

@0xSero Does mission mode work in the desktop app or? Can’t seem to find it in mine, works from the cli though

Here's why I shill Droid 24/7 ---------- Today Droid single-handedly: 1. Published a REAP of GLM-5 in FP8, there's a reason no one else has done it DSA is still very new: huggingface.co/0xSero/GLM-5-R… 2. Found and Fixed an upstream issue with VLLM + DSA + Hopper where GLM-5's kv-cache would need to recompute and spend 20x the time needed, fixed. 3. Created multiple working quantisations on it's own, it tried exl3 and autoround but both failed so resorted to GGUF (autoround 3 bits doesn't work on ampere) huggingface.co/0xSero/GLM-5-R… 4. Implemented github.com/0xSero/turboqu… within 24 hours of the research paper coming out, tested it across 5090s, 3090s, H100s, and B200s 5. Has been distilling larger models into LoRA to help me test arxiv.org/abs/2505.21835 and it got an 80% prune to be semi-coherent again. 6. Helped my find research papers, clean up slop with the human-writing skill. 7. Got BYOK working with Anthropic, ZAI, Kimi, MiniMax, OpenAI working in Cursor github.com/0xSero/factory… 8. Helped me Implement blog.comfy.org/p/dynamic-vram… 's dynamic loading, only works on a tiny model, but still. ------- I only have to check in on it every 30-45 minutes (I am talking all 8 of my sessions) the thing will run for 16 hours with like 0 prep All this while I am mostly focused on my actual job and tweeting 24/7 Keep in mind each one of these experiments is running on a different server, with different constraints, like I don't understand how I can get such good results here. --------- I love novelty. Which is why I jump around and talking about all these different tools. I have used all of these harnesses and messed around with every feature. I keep coming back to this, and I keep shilling it because I sincerely wish others get to experience this.

@GrahamJCampbell They were paying a flat subscription so how could they spend more? What are you classing as “spend sufficient money”? X user no longer gets X limit they was paying for, its a fair market and a valid reason to not continue paying for X thing lol

Hot take. Everyone cancelling their Claude subscription because the subsidisation is now less were shit customers who were never going to spend sufficient money anyway. Google learned this the hard way attracting the bottom of the market.

tmux + xterm + node-pty = horrible scrollback artifacting. I struggled all week with maxxed claude + codex + gemini. all failed. I realized the solution was to drop tmux and persist node-pty in a sidecar. None of the agents even proposed this approach. github.com/collaborator-a…

@0xSero I think I have all the above apart from BYOK lol Claude opencode and codex work with your local auth though

Please list all the apps you know of this category that's not already on the list: 1. Cursor Glass 2. Factory Desktop 3. Codex App 4. OpenCode App 5. Claude Desktop 6. CMUX I am looking for this mythic ADE: 1. Has browser 2. Has filesystem 3. Has BYOK 4. Has good ui