If you like the idea of using Glass but don’t want to wait, try @usehamster It’s the multiplayer self-improving, self-healing workspace we’ve been working on for more than a year that streamlines the full SDLC And it’s S-tier - full direction loop for the whole team - discovery and alignment across living artifacts - cloud delivery across briefs and initiatives - shared method and skills + a curated library of methods and skills to one click install - routines for automating work across the team - self-writing blueprints describing product that auto update as your team ships - self-healing context graph with team decision intelligence and intent lineage to answer what was built, when, why and by who and who made the decisions Built with love by the team who built @TaskmasterAI

"which quant should I download?" is a question you may never have to answer again the team @HamsterResearch has figured out how to kill it with pMLX. download once at full precision (bf16) and the engine re-fits it to your machine on the fly, based on the job you give it tell it two things: how much context you need, and the slowest speed you'll accept. it reads your mac and picks the quantization plus how many experts stay in ram vs. stream from disk. no need to download a smaller quant or figure out which quant fits your hardware. same model for every use case and the config changes based on what you need I ran this on my own M4 Max and Qwen3.8-Flash-Next splits into two configs. short context, under 32k: full bf16, ~85% of experts in ram, fast at full precision since it fits my ram long context, 64k to 256k: keep ~70% bf16 and stream the rest from SSD at 22 tok/s, or drop to q8 and get 40 tok/s. I pick per task and the model itself never changes. there are many possibilities since I have the ram to spar - if i need speed, Q4 100% resident (74gb ram @ 62 tok/s) - if i need balance, Q8 95% resident (76gb @ 38 tok/s) - if i need accuracy, bf16 70% resident (96gb @ 16 tok/s) given whatever RAM you have (16/32/64/128/256 GB) + your context + your min speed, the engine picks the precision (bf16→q8→q4) and the expert-residency/paging split that fits your box and maximizes quality & speed last thing to optimize is speed. there are so many things we want to power with open models at @usehamster and these 180b-300b class models have the potential to play a big role in that

More Qwen3.8-Flash-Next breakthroughs, releasing soon in a new tiered model and supported by a custom MLX engine built by the @HamsterResearch team - you only ever download the bf16 tiered model (upcoming) - pMLX can quantize it up/down bf16/q8/q4/q3 on the fly - 130-150 tok/s aggregate decode (crucial for sub agents) - adjustable n-gram streaming from nvme - adjustable experts residency ratio - prune/REAP experts on the fly - powered by pMLX hundreds of config possibilities self-tuned from your hardware examples for code-editing (which this decodes 1.5x faster): - q4 @ 32k context, 95% experts in ram, n-gram in memory, 87 tok/s burst, 58 tok/s sustained (72gb memory), 149 tok/s aggregate - q4 @ 256k context, 95% experts in ram, n-gram in memory, 87 tok/s burst, 50 tok/s sustained (74gb memory), 130 tok/s aggregate - q3 @ 32k context, 95% experts in ram, n-gram in memory, 111 tok/s burst, 74 tok/s sustained (54gb memory), 190 tok/s aggregate lowest memory footprint config: 12gb 🤯 - q3 @ 8k context, 30% experts in ram, n-gram on disk, 16 tok/s burst, 10 tok/s sustained (12gb memory), 29 tok/s aggregate launch targets achieved, now packaging it up for you!

the highest value thing a context graph stores is why your product ended up this way llms trend toward the average. that's fine until you're implementing, because general models fill in the corners with decisions that are directionally correct and subtly wrong, and those come back to bite you in the ass weeks later when you're debugging something you never chose this works for disposable discovery work that eventually leads to decisions but if you're shipping production code this way, you'll inevitably decide planning ahead is the fix but there's a catch. if you ask ai to lay out the options for you and it hands you four generic choices, you just moved the averaging one step earlier in the process this is a question hamster asked me this morning while refining a brief. the options aren't generic. they came from decisions we already made and recorded. they're things that matter to us that it knows to surface because the context graph contains these types of decisions. your past decisions are the highest leverage context you have. everything else is just documentation.

Not relying on mcp tools to collect context or keep it in sync is a wonderful feeling

every time you start a new claude code session you spend the first 10 minutes re-explaining your project "here's the stack. here's what we're building. here's what was decided last week. here's the file structure. DON'T TOUCH THIS PART WE JUST REBUILT IT" but it's not just your own context. your teammates shipped 3 new things since yesterday, two architectural decisions changed and a dependency got swapped out. you need to capture all of that too, or claude is building from a stale snapshot you won't remember everything. so claude compiles context from scratch, misses things, and pulls in a direction that might completely miss the point. now multiply that by every dev on your team, every session, all day literal coordinated drift what we do instead: the .hamster folder syncs decisions, plans, and context to the filesystem automatically. claude reads it and just knows. no re-explaining. no stale snapshots. everyone's always got a fresh .hamster folder which updates in real-time as the rest of the team (or you) ship decisions, tasks, briefs, methods, skills, etc try it. you'll love it → @usehamster

everyone on your team maintaining their own CLAUDE.md is a symptom, not a solution. the root cause is that shared context has nowhere to live and evolve as your team makes decisions. you can't fix it with naming conventions or a better folder structure. decisions need a home. not a static file someone remembers to update. a workspace that accumulates as the team works. when a decision changes, it changes once, and every agent on the team pulls from that. there's no need to keep the markdown files up to date or copy-paste anything anywhere. claude just knows. your PRDs, your tasks, your skill files, your CLAUDE.md are all projections of the same underlying decisions. if the decisions live somewhere real, everything else stays aligned. if they stay up to date automatically, the rate of progress can explode because the synchronization overhead disappears. that's what we've been building into hamster. decisions get a home. blueprint holds your product's current state. methods encode how agents should work. the projections stay aligned because the source of truth is real. dm if you want to try it GA reeealllly soon!

we've shipped huge features in taskmaster v0.37 🚀 → real-time PRDs, team plans, synced tasks 🐹 → GPT-5, @lmstudio, @Zai_org GLM → @claudeai plugin + @aisdk v5 → smaller MCP tool footprint → redesigned init ux → enterprise proxy + more follow + retweet + bookmark 👀👇

multiplayer chat is the next great collaborative ui we use it for teaming up to draft and refine PRDs and briefs that then get executed by background agents

know incredible devs? paying $5K for successful referrals looking for: - senior TS engineers to own core pkgs for @usehamster - lead integration specialist to help me build context graph from tools we use daily - lead ML specialist w/ exp in custom model dev dms are open! 🐹