AI Daily Digest · 2026-06-03
🔬 New AI Craft
1. UltraCode-Shim — Give Any Model Claude Code's ultraCode Mode
UltraCode-Shim is a minimal local proxy that exposes Claude Code's hash-anchored editing and tool orchestration (ultraCode mode) to any LLM via a single config.json. Unlike the Agent Plan approach you use daily, ultraCode focuses on fine-grained, deterministic edits — this shim decouples that capability from a specific model, letting you switch providers while keeping the same toolchain.
https://github.com/OnlyTerp/UltraCode-Shim
2. How kapa.ai Indexes Images for RAG
A detailed engineering post on adding image understanding to RAG pipelines — covering visual feature extraction, cross-modal embedding alignment, chunking strategies, and retrieval evaluation. Goes beyond basic vectorize-and-search to tackle real-world multimodal indexing challenges.
https://www.kapa.ai/blog/how-we-index-images-for-rag
3. Bringing Up DeepSeek-V4-Flash on AMD MI300X
A hands-on deployment guide covering ROCm environment setup, custom operator optimization, and inference tuning for running DeepSeek-V4-Flash on AMD's flagship MI300X accelerators.
https://fergusfinn.com/blog/deepseek-v4-flash-mi300x/
🛠️ Tools & Tips
1. Vigils — Local Control Plane for AI Agents
A Rust + Tauri desktop app that gives you real-time visibility into agent behavior, approval gates for sensitive actions, and secret isolation — built for teams worried about agent autonomy risk.
https://github.com/duncatzat/vigils
2. Prompt Cache Skills — Drop-in Caching for AI Coding Agents
Point this at your agent harness repo and it auto-patches prompt caching; reduces repeated inference across multiple prompt patterns.
https://github.com/OnlyTerp/prompt-cache-skills
3. nbd-vram — Use NVIDIA GPU VRAM as Swap on Linux
A Linux kernel module that exposes GPU VRAM as a block device for swap — useful when local inference runs into VRAM limits without buying new hardware.
https://github.com/c0dejedi/nbd-vram
⭐ Open Source Highlights
1. odysseus (32901⭐) — Self-Hosted AI Workspace
A unified platform for managing multiple AI tools and models under one roof.
https://github.com/pewdiepie-archdaemon/odysseus
2. Mellum2 — JetBrains Open-Sources Its AI Workflow Model
JetBrains released Mellum2 as open-source, a fast model optimized for AI-assisted development workflows including IDE completions and code analysis.
https://blog.jetbrains.com/ai/2026/06/mellum2-goes-open-source-a-fast-model-for-ai-workflows/
3. Paseo — Beautiful Open-Source Coding Agent Interface
A polished UI for interacting with coding agents, supporting multiple models and custom workflows.
https://github.com/getpaseo/paseo
📰 Industry News
1. Trump signs downsized AI order after weeks of reversals
2. California's university system went all in on AI, now it's tearing itself apart
3. Archestra raises $10M for open source AI security
🚀 Major Releases
1. Microsoft launches MAI-Code-1-Flash, a new code-focused model
2. Anthropic expands Project Glasswing, scaling AI safety research
