Phonon on Ollama: Gradium's tiny text-to-speech family, first look
Gradium's phonon is on the Ollama library as of 24 October 2026 — a text-to-speech family small enough to run on any Mac, with a licence split worth knowing before you build on it.
The Inpherio blog
Everything we've written about engines.
Listen to the latest in local AI. New models, GPUs and how-tos — every post is also read aloud, so you can catch up on the commute, at the gym or with your phone locked. Change the speed, pick up where you left off.
Accessibility: every post has a spoken version with a keyboard-friendly player, plus clear headings and a table of contents.
21 posts · page 2 of 2
Gradium's phonon is on the Ollama library as of 24 October 2026 — a text-to-speech family small enough to run on any Mac, with a licence split worth knowing before you build on it.
NVIDIA's compact vision-language model for local coding, reasoning and tool use has landed on the Ollama library in two very different sizes — one that fits a 12GB GPU, one that realistically needs more than 24GB of VRAM.
Kev is a new "decision model" on Ollama — it doesn't chat, it scores answers to structured questions in a single forward pass. Here's what each size needs and how to call it.
vLLM is the engine you pick when you need throughput, but the install is where most people give up. Here's the whole path: pip vs Docker, the Windows reality, a working `vllm serve` command, and a five-minute benchmark to prove it.
Google DeepMind's EmbeddingGemma 2 is now on Ollama — a 740M-parameter multimodal embedding model that puts text, code, images, video and audio into one shared vector space, small enough for a laptop.
Cloudflare's Clef Flash has landed on Ollama — a 9B model that doesn't chat, doesn't write, and just makes decisions, fast. Here's what it is, what hardware it wants, and how to run it.
A new kind of local model has appeared on Ollama: one that doesn't chat at all, but returns probabilities for decisions you define. Here's a practical first look at Strands Decider.
Bespoke Labs' nimble-v3 has landed on Ollama. It's a 9B model fine-tuned from Qwen3.5-9B that doesn't chat — it routes, checks and scores, returning probabilities instead of text.
One is a point-and-click desktop app, the other is a terminal command — and people are thoroughly confused about which does what. Here's the clear answer, plus why "LM Studio is slow" is usually not LM Studio's fault.
Have a GPU, a Mac or a spare PC?
Run open models for people who need them. You set the price and keep everything but the 15% fee.
Become a hostBuilding with AI?
OpenAI-compatible. Priced by the people who run the hardware.
Start free See the marketplace