
#microsoft / experimenting / designing copilots / posting about AI, design and innovative products+ideas
Joined June 2011
- Tweets6.7K
- Following955
- Followers1.9K
- Likes2.3K
Tash Ahmed retweeted
🔑 Free model tiers move every week, and each one still wants its own key, its own base URL, and its own outage.
Meet FreeLLMAPI by @tashfene: a free, MIT-licensed Mac menu-bar app that runs a local OpenAI-compatible router. Version 0.16.19, released November 4, 2026. The GitHub repo was created on April 21, 2026, so this is a new OpenNavo listing and a new Homebrew cask, not a first release.
➕ Add the new provider. 0.16.19 adds FreeInference to the provider picker and the key import flow. It supports tested Qwen3.6 35B chat, including streaming, and BGE-M3 embeddings with 1,024-dimensional vectors.
🔎 Check the key for real. The release validates upstream model identities and embedding shapes, keeps normal routing errors and failover, and does not treat FreeInference's public model roster as API-key validation. Real requests establish key health.
🧭 Point clients at one port. The README puts those providers "behind a single `/v1` API." "Keys are stored encrypted." The router "picks the best available model for each request, falls over to the next provider when one is rate-limited, and tracks per-key usage so you stay under every free-tier cap." The menu-bar app runs that router and dashboard locally, and the only credential you need to start is the unified API key from the tray popover.
Why it's worth watching:
✅ You already point Codex, a script, or another OpenAI-compatible client at more than one free-tier provider and want one local base URL.
It is not a new app, and it is not a hosted chat product. The app is the router. README limitations: no frontier models, variable latency, no SLA, and the endpoint gets weaker late in the day as top models hit daily caps, then resets at UTC midnight.
FreeInference offers a shared daily research allowance. It logs prompts and responses for research, so do not send sensitive data. Free installs receive new catalog entries through the unchanged 30-day release gate. Premium receives the new catalog the same day. README pricing is $19/year or $49 once. The router itself stays MIT-licensed and fully free. Premium is only the live feed.
Needs macOS 12 Monterey or later. Choose the arm64 build for Apple Silicon or the x64 build for Intel.
Discover FreeLLMAPI on OpenNavo:
opennavo.com/apps/freellmapi
#MacApps #MacDev #LLM #OpenSource
I've been training the most natural realtime Urdu voice model at home 👇
Most tools I've tried haven't felt very natural in Urdu. I wanted something that sounds closer to how we actually speak Urdu.
Been paying a lot of attention to the little things, the rhythm, the pauses, how a sentence sounds when someone's just having a conversation. And all of this is being trained on my DGX Spark cluster sitting at home.
2
13
1,131
I'm building an AI voice that actually speaks Urdu naturally.
It's training right now on a small cluster of DGX Sparks in my home. Here's where it's at 👇🔊
7
6
686
Early results. Still training, but listen to the urdu lehja.
Every big AI speaks Urdu like a tourist. This one is starting to sound like home.
1
2
354
I'm speaking on a brilliant panel at #DesigningFutures in London on 5th November.
All profits go to Children of the Mekong, a charity that helps children in Southeast Asia access education. So you'll be doing something good just by turning up.
Join us: luma.com/4lkgf2hc
1
1
360
So good! Can't wait to experiment with this.
$52M seed round. 15M+ users. But this is what we're most proud of yet.
Introducing Drama 3: the most emotive and controllable TTS model in the world
And yes, it's open-weight.
You can now control how even a single word should sound, change its tone, and shift your voice mid-sentence.
Uncle Roger was impressed too.
Try it here:
s.fish.audio/rqh0l8
Comment “DRAMA” within the next 48 hours for 1 month free.
If you're an enterprise your first month is on us:
s.fish.audio/qz9r0t
1
271
Used rea (reverse engineer anything) to build a GTM tool.
It's wild! 🤯 It can get prospects, verify and manage leads, warm up emails and set campaigns and has an AI AGENT to do all that:
github.com/tashfeenahmed/kin…
2
8
347
Been working on some use cases for Word using decision model - excited to try out Microsoft’s Decision-1 this week!
1
199
Tash Ahmed retweeted
Introducing Microsoft-Decision-1, our new model for fast decision-making.
It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality.
We’re already testing it across Microsoft for everything from incident response and quality control to scientific discovery.
Tash Ahmed retweeted
🚨 LA MAYORIA NO CONOCE ESTE REPO
Decenas de proveedores de IA gratis.
Si uno se cae o se acaba la cuota,
pasa al siguiente.
Se llama FreeLLMAPI.
🤯我靠!再也不用为 API 额度发愁了!开源神器 FreeLLMAPI:把 30+ 家免费 LLM 额度叠成一条 OpenAI 兼容接口了!
理论上限约 74 亿 Token/月,叠加各家免费档,实际能用多少取决于你加了哪些 Key、各家限流和条款
具体操作步骤:
1️⃣装本地网关(推荐 Docker 一行)
curl -fsSL freellmapi.co/install.sh | bash
打开 http://localhost:3001
Windows 可下桌面版:github.com/tashfeenahmed/fre…
2️⃣去各家官网自己注册免费档,把 API Key 贴进后台「密钥」页
常见:Google Gemini、Groq、Cerebras、Mistral、OpenRouter 免费模型、Cloudflare、Hugging Face、智谱 Z.ai、NVIDIA NIM、魔搭 ModelScope 等(具体名单以站点目录为准)
3️⃣拖一下「回退链」顺序:这家限流了自动切下一家,并按 Key 记账,尽量不超各家免费上限
4️⃣复制页面上的统一 Key(freellmapi-...)
任何 OpenAI 兼容客户端改两处即可:
base_url = http://localhost:3001/v1
model = auto(也可 auto:fast / auto:smart 或指定模型)
5️⃣Cursor / Claude Code / Cline / Continue / Aider 都能接,仓库里还有一键 setup 脚本
注意事项:
🔹路由器本身开源免费、密钥本地加密;实时目录更新是 Premium(约 $19/年),免费版目录会滞后
🔹只用各家官方免费档,遵守 ToS,仅限个人实验
🔹仓库:github.com/tashfeenahmed/fre…
🔹目录:freellmapi.co
#FreeLLMAPI #白嫖Token #开源AI #LLM #Cursor #本地部署 #OpenAI兼容 #梁文峰
Tash Ahmed retweeted
FreeLLMAPI turns scattered free LLM tiers into one OpenAI-compatible API.
Self-host it, add your keys, then let it handle routing, quotas, and failover. Here's how it works.
why pay for 10 ai tools when you can use their free tiers? 😳
found another open source FREELLMAPI tool
one api for multiple free ai providers
google, groq, cerebras, mistral, openrouter, nvidia + more
how to use:
→ download the desktop app
→ add your provider api keys
→ choose your favorite model
→ start cooking
it can automatically switch providers when one hits its limit
works with claude code, codex, gemini cli, cursor and cline
one api → hundreds of free endpoints 😭
another open-source hit worth bookmarking
github.com/tashfeenahmed/fre…
Tash Ahmed retweeted
A 35B language model running on an iPhone using only 1–2.5 GB of peak memory.
No cloud. No remote server. No desktop GPU.
Today, we’re open-sourcing Edge0 — a framework for running large AI models fully on-device.
I found three GitHub repositories that transform an AI from a conversationalist into a working system:
1. FreeLLMAPI ★ 18.4k
Consolidates free limits from 29 LLM providers into a single OpenAI-compatible API
The router automatically selects an available model and switches if one reaches its limit
2. MultiModels MCP ★ 18
Allows Claude Code to delegate tasks to GPT, DeepSeek, GLM, OpenRouter, and local models.
One model builds a plan.
Another checks the code.
A third performs cheap routine work
3. Netlify MCP ★ 56
Gives the AI agent access to Netlify
You can ask it to create a project, deploy a website, set environment variables, and manage the infrastructure with a standard prompt
The main idea:
A powerful AI assistant isn't one big prompt
It's a system where models delegate tasks to each other, connect to services, and perform work under your control
I've left links to all three repositories in the reply ↓













