
BlockedPath
6.7K posts

BlockedPath
@BlockedPaths
Building personal AI infrastructure. Local LLMs, coding agents, open-source models. Real hardware, real constraints. Learning in public.
Maryland 参加日 Nisan 2009
419 フォロー中756 フォロワー

@Rakib_Web3 @nahid_pro09 No its not free, you have to put your providers in
English

You can run a truly FREE "claude code" environment using a brand-new open-source agent that aggregates 24 different ai providers😳
most people are burning money on expensive subscriptions, but this tool connects you to free trials, budget-friendly options, and free-tier models without the usual overhead
takes 5 minutes to install
what you get inside this agent:
-> integration with 24 ai providers to pool your trial credits
-> support for $1 commandcode go plans for low-cost scaling
-> access to free trials from kiro, kilocode, cline, antigravity, and mistral
-> native support for free models like deepseek 4 flash via opencode
-> full compatibility with the claude code ecosystem, including skills, plugins, and mcp servers
-> full setup guide (tau agent implementation):
step 1: access the repository
->> go to the official github page at github.com/AbdoKnbGit/tau
step 2: install your environment
->> clone the repo and follow the instructions to set up your local agent workspace
step 3: configure your providers
->> plug in your choice of the 24 supported providers to maximize your free trial and budget tiers
step 4: launch your claude code environment
->> activate the mcp server and plugins to mimic a professional claude code workflow for free
->> already paying for high-cost coding assistants? skip straight to step 3 and replace your paid tokens with these free and budget alternatives
this is the most effective way to build a high-capability coding assistant while keeping your costs at zero
bookmark this before the closed testing slots fill up completely

English

@ahmedehab_01 I would try omp first, its built on Pi but has features like autoresearch, and alot of providers.
English

@BlockedPaths Thanks a lot!
I heard a ton of cool stuff about Pi, and omp seems to be Pi + some cool features, probably will try omp first.
English

@william_arin @doregex I believe dario or elon said they are bout 3 months behind. When we release a new model they distill their modes with it, Kinda cheating. Ive built a harness with deepseek.
English

@doregex @BlockedPaths Yes, but I'm not sure if they will catch up that quickly. Last december it felt like Chinese models lagged 4-5 months behind. But GPT 5.0 in september already understood intent very well, and 9 months later Chinese models are still not at this level I think.
English

@onesuitee @BlockedPaths Ask to grok cli to configure it directly, it will update its config file, ask you to restart and voilà, you get it working.
English

@Amank1412 Something is off with their token usage, I went through an insane amount of tokens on a simple task. It mad Claude look generous.
English

@blueiswise @ZixuanLi_ I had this problem as well, it was way to quick.
English


Thanks for all the feedback. GLM-5.2 will begin rolling out to all Coding Plan users in 3 hours.
Zixuan Li@ZixuanLi_
Help us shape the next GLM release: what should we prioritize most?
English

@usr_bin_roygbiv @usr_bin_roygbiv we have the same brain omg lmao
English

you mean droid missions and ralph loops from a fucking year ago?
ghuntley.com/ralph/
Skiipy@Skiipy88
@usr_bin_roygbiv Codex /goal is the best feature released this year. Worked 50hrs unsupervised on a project Opus simply gave up on. Only thing is to give the constraint to not work aimlessly, otherwise it would keep looping for nothing because 1 part of the goal was needing intervention lol
English

【GLM-5.2がCoding Planで利用可能に】
Z. aiは、GLM Coding Plan利用者向けにGLM-5.2を提供開始した。Claude Code、OpenClaw、Clineなど、カスタムモデル設定に対応するコーディングエージェントから利用するモデル。
GLM-5.2は1Mトークンのコンテキストに対応。Claude Codeではglm-5.2[1m]を指定し、自動圧縮ウィンドウCLAUDE_CODE_AUTO_COMPACT_WINDOWを1000000に設定することで1Mコンテキストを有効化できる。
公式Xでは、API(外部からモデルを呼び出す窓口)とChatbot(Web上の対話画面)の提供、MIT Licenseの下でのオープンソース公開を翌週予定としている。

日本語


Maybe it’s just me, but I’m getting tired of seeing the same AI terminal launched over and over again.
Warp.
Conductor.
Superconductor.
CMUX.
Ghostex.
And another one next week.
Every demo looks nearly identical:
“Look, I can run multiple agents in panes.”
At some point this stops being innovation and starts being a UI clone war.
English

@therealZpoint Had me confused af lol. You owe me some tabs if herdr is no bueno hahah
English

@BlockedPaths We is just a stylistic thing. Like our public pronoun on X. Although we find that pronoun game super gay, too. lol
English

@therealZpoint I stared herdr, im just fatigued of the abundance of software
English

@BlockedPaths We never launched any of those. Only herdr. That's enough.
Also, we're on Linux.
And any app starting with "super" is suspicious.
Btw, Tabs in terminals are super gay.
English

Apple's Foundation Models framework got much bigger in Xcode 27.
I added runnable labs for:
- Private Cloud Compute + reasoning levels
- Image input
- Dynamic profiles
- Explicit tool calling
- Model routing
- Evaluations
github.com/rudrankriyam/F…
English

@Maximus_W3 Here is my extension for pi, hopefully it works for you as i just made it for you and posted it on github
github.com/BlockedPath/pi…
English

@BlockedPaths I checked access as soon as the model was released. Perhaps there is now access to the API through third-party methods
English

@mizanyx @usr_bin_roygbiv I jailbreak every chinese model so refusals are rare for me.
English

@usr_bin_roygbiv @BlockedPaths Zero refusals?
One of the first noticeable innovations from Moonshot was making their models opinionated instead of sycophantic.
Your prompting is too vanilla if the model never said "no" to you.
English

@Maximus_W3 I managed to get it, what problem are you having? I had to build some extensions to get 5.2 in my harness though
English









