BotPsy pairs an on-device assistant with a small agentic model
The Hugging Face post describes a personal assistant built around a language model under one billion parameters.
Stories per day · 30 days
3 RSS
Relatedopen-weights7liquid-ai5agents4huggingface4language-model4lfm2.53lfm33multimodal3image-editing2mixture-of-experts2open-source2spark2apache-2.01apple1attention1benchmark1bishield-tee1block1
35 stories, newest first · one card per story
Stories tagged #on-device
GAAC losslessly compresses BF16 KV caches about 1.41×
On NVIDIA DGX Spark, 64K-token decode attention sped up 1.19–1.27× and resident KV memory fell about 26%.
MoI-Bench scores mobile image generation and editing gaps
The suite has 64 subtasks and 1,150 samples, with MoI-Eval scoring instruction following, quality and detail preservation.
Liquid AI releases open-weight multimodal LFM3 models around 2B
LFM3-1.7B-Base was pre-trained on 22T tokens, with a 256K context after mid-training.
PREEN predicts editor presets for exemplar color transfer
The model regresses 29 smartphone-editor sliders, runs at sub-second on-device latency, and leads PST50 fidelity metrics.
Hob pitched as open-source on-device alternative to Claude and ChatGPT
The Hacker News post has 20 points and 7 comments and gives no feature or release details.
Block releases Dessa, a local full-duplex speech model
Dessa is not publicly available, and Block has asked safety researchers to test it first.
2
Gradium open-sources Phonon, a 31M-parameter on-device TTS model
The release is licensed under Apache 2.0; the article body was not available beyond the title.
4
Liquid AI releases on-device omni-modal model LFM3-Omni-2B
The 2.17B checkpoint accepts text, images, video and 16 kHz audio, and can reply with text, 24 kHz speech, or both.
Liquid AI releases LFM3-VL-2B on-device vision-language model
The 1.92B model takes text, images, and video, reasons step by step, and uses the LFM Open License v1.0.
1
XHToken releases Spark-X2.6-4B GGUF for on-device use
The 4B model keeps the Spark-X2.5 architecture and claims coding and agent gains from post-training alone.
XHToken publishes Spark-X2.6-4B on-device model weights
The 4B model keeps the Spark-X2.5 architecture and adds none, low, and high thinking levels.
Muse Glimmer 30B released as an on-device agentic model
The Apache 2.0 model takes images, runs at about 4-bit on one GPU, and claims to beat same-size peers.
Meta and Apple team up to make Muse AI models run better on Macs
Meta released Muse Spark Open, an 816B open-weight model tuned with Apple to run on Mac Studios.
BiShield-TEE: On the (In-)Security of Unilateral Weight Obfuscation in On-Device TEE-Shielded LLM Partition
Unilateral TEE weight obfuscation for on-device LLMs is breakable; BiShield-TEE hides the key among ambiguous solutions.
-oOdUn61bDRSykk4H951SoDOHzqoOe1.png&w=828&q=75)
LFM2.5-24B-A2B: Built for fast on-device agents
Liquid AI released open-weight LFM2.5-24B-A2B, a 24B MoE with 2B active parameters for on-device agents.
Liquid AI Releases LFM2.5-24B-A2B: A 2B-Active MoE for On-Device Agents
Liquid AI released LFM2.5-24B-A2B, an open 24B MoE with 2.45B active parameters for local agents.
Google DeepMind and Singapore Government collaborate to build fine-tuned AI model for scam detection
DeepMind and Singapore's GovTech fine-tuned on-device Gemma to detect scams without cloud message scanning.
4
Run fast, private agents on-device with LFM2.5-24B-A2B
LiquidAI’s LFM2.5-24B-A2B is pitched for fast, private on-device agents.
Google’s AI note-taking app transcribes your meetings completely offline
Google launched AI Edge Foresight, a free Mac app that transcribes and summarizes meetings fully offline.
Google releases a new local-first Granola competitor
Google launched offline Mac note-taker AI Edge Foresight, using on-device EmbeddingGemma 2 and Gemma 4.
4
EmbeddingGemma 2: an open, lightweight multimodal embedding model
Google DeepMind released EmbeddingGemma 2, a 740-million-parameter open multimodal embedding model for on-device use.
7
openbmb/MiniCPM-V-4.7-35B-A3B — new model trending #29 on Hugging Face
OpenBMB released MiniCPM-V 4.7 35B-A3B, its first on-device mixture-of-experts vision-language model.
openbmb/MiniCPM-V-4.7-35B-A3B-gguf — new model trending #30 on Hugging Face
OpenBMB released GGUF weights for MiniCPM-V 4.7 35B-A3B, an on-device multimodal MoE model.