惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Blog — PlanetScale
Blog — PlanetScale
Vercel News
Vercel News
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
量子位
Y
Y Combinator Blog
IT之家
IT之家
博客园 - 聂微东
L
LangChain Blog
爱范儿
爱范儿
H
Help Net Security
GbyAI
GbyAI
F
Fortinet All Blogs
B
Blog
Microsoft Security Blog
Microsoft Security Blog
罗磊的独立博客
C
Check Point Blog
博客园 - 三生石上(FineUI控件)
小众软件
小众软件
D
DataBreaches.Net
Last Week in AI
Last Week in AI
WordPress大学
WordPress大学
B
Blog RSS Feed
酷 壳 – CoolShell
酷 壳 – CoolShell
宝玉的分享
宝玉的分享

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - brxs/slipmate: An AI DJ application
ttoinou · 2026-06-15 · via Hacker News - Newest: "AI"

SlipMate — generative DJ

Your generative DJ mate — a DJ instrument for real-time AI music.

Buy me a coffee

Two locally-running model decks, steered by text prompts and mixed like vinyl: three-band EQ, one-knob Color FX, a crossfader, headphone cue, and full Pioneer DDJ-FLX4 control. The live decks run on Magenta RealTime 2; generated pads and finished tracks come from Stable Audio 3. See docs/ROADMAP.md for how it got here and docs/adr/ for the architecture decisions.

Requirements

  • Apple Silicon Mac (MLX backend)
  • uv
  • ~13 GB disk for model weights (downloaded on first setup: Magenta ~4.5 GB for both deck models, Stable Audio 3 ~8 GB including the medium track model)
  • A Chromium-based browser (the app leans on Web Audio worklets and Web MIDI; it is developed and verified against Chrome)
  • Optional: a Pioneer DDJ-FLX4 for hardware control and its headphone jack

All common tasks live in the justfile — run just to list them.

Setup

just setup   # backend deps, all model weights (~13 GB), frontend deps + build

Magenta models land in ~/Documents/Magenta/magenta-rt-v2 (override with MAGENTA_HOME): both deck models, the default mrt2_small and the heavier, higher-quality mrt2_base, selectable per deck in the UI — the app warns when the combined selection looks tight for your RAM. Stable Audio 3 — generated pads and tracks — is cloned to ~/Repos/stable-audio-3 (override with SA3_MLX_HOME; an existing checkout is reused) and its weights are pre-warmed so no request ever pays for a download; just setup-sa3 re-runs that half alone.

Run

Then open http://127.0.0.1:8000 — add style targets to a deck's pad, hit play, blend targets by dragging the cursor (or the dots themselves, to cluster them), and ride the crossfader between decks.

  • Mixer — per-deck volume and Hi/Mid/Low EQ, crossfader, and Record, which captures the master bus to a downloadable WAV. The health row shows the stream buffer, underrun count, and generation speed.
  • Color FX — one knob per deck over a chosen effect: Filter (bipolar LPF/HPF), Dub Echo, Space, Crush, Noise, Sweep. The knob's centre/zero is a bit-exact bypass (ADR-0008).
  • Freeze loops — capture the last bars of a deck into one of four loop slots and hold the moment on air while you re-steer the model underneath; loops are session-only by design (ADR-0009).
  • Beat detection — each deck shows its detected BPM behind an honesty gate (a dash rather than a wrong number); with a confident tempo the Dub Echo syncs to the beat and freeze captures quantise to whole beats (ADR-0010).
  • Deck-to-deck style sampling — one press puts "the sound of the other deck, right now" on a deck's style pad as a blendable target; sampled targets are session-only by design (ADR-0011).
  • Crates — save a deck's pad + Color FX as a named preset, browse the crate from the FLX4's rotary, and load onto either deck mid-set; export/import as JSON for backup and sharing.
  • Master housekeeping — a limiter on the master (the meter, the recording, and the phones all hear the limited signal; its gain reduction shows in the mixer) and per-channel auto-gain Trim that levels decks of different loudness, with a manual override.
  • Headphone cue — hit a channel's Cue, ride the Cue mix knob between cue and master, and pick a Phones out: any output device the browser can reach, or the FLX4's own headphone jack, which is fed by the backend over USB (ADR-0007).

Settings (pad arrangements, volumes, crossfade) persist across reloads. Shortcuts: A/B focus a deck's style-target input, X focuses the crossfader.

For frontend development: just dev-backend in one terminal, just dev-frontend in another (the Vite dev server proxies /ws to the backend).

Hardware control (Pioneer DDJ-FLX4)

Plug in the FLX4 and click Connect MIDI (Chrome asks for MIDI with SysEx; plain MIDI works too, minus position sync). Mapped controls:

  • Play/pause, channel faders, three-band EQ, crossfader
  • Channel CUE buttons (headphone cue) and the transport CUE button (deck prep: prime a stopped deck off-air, stop a playing one)
  • SMART CFX knob — Color FX amount; hold SHIFT to sweep the style pad instead
  • PAD FX pad bank — select the deck's effect (re-press toggles it off); HOT CUE pads pick style targets; SAMPLER pads freeze loops (SHIFT + pad clears a slot)
  • HEADPHONES MIX knob — cue mix
  • Browse rotary + LOAD buttons — highlight a crate preset, load it onto deck 1 or 2

Knob and fader positions sync from the hardware on connect, and the LEDs mirror app state. The measured byte map lives in docs/midi-ddj-flx4.md.

Verify

  • just test — backend pytest + frontend vitest
  • just lint — format check, ruff, eslint, tsc
  • just check — both of the above; what a PR must pass
  • just verify-stream / just verify-ui — e2e against a running server (UI e2e needs Playwright Chromium once: npx playwright install chromium in frontend/)
  • just verify-worklets — the audio-worklet module graph loads in real Chromium (self-contained; jsdom executes none of the worklet code)
  • Hardware behaviour is verified by a human against the checklists in docs/ (m7-, m9-m10-, m12-hardware-checklist.md)