Are you the maintainer?
Claim this page →A compact multimodal model that processes text, image, and audio inputs with a 128K token context, supporting OCR and chart and table understanding.
01 / About
Phi-4 Multimodal processes text, image, and audio inputs in a single model with a 128K token context length. It supports OCR along with chart and table understanding for multimodal document tasks.
Reach for Phi-4 when you need lightweight multimodal with speech: it handles text, vision, and speech in a compact model, making it a great fit for on-device agents.
02 / Discussion CREDIBILITY-GATED
Reading is open to everyone. Only verified humans or builders at GitHub B+ can post or rate — every comment carries its author's credibility.
03 / Related
ToolControl browsers programmatically with LLM agents through a high-level, LLM-friendly API.
ToolJuliusBrussee
An experimental LoRA fine-tune of Gemma to speak caveman-mode natively.
ToolJuliusBrussee
A TypeScript implementation of the caveman compression engine.
Toolegorfedorov
Tracks token usage, blocks redundant reads, and supports .contextignore and budget alerts.
ToolNodeNestor
A proxy plugin that rolls context compression past 100K tokens.
Toolnadimtuhin
Restructures CLAUDE.md and docs for roughly 90% token savings with a CLI audit and compress.
04 / Build
Browse the catalogue for harnesses, tools, and blueprints — each scored on real GitHub credibility.
Browse the catalogue