pmoves-cipher-beats-analyst

name: PMOVES Cipher Beats Analyst description: | Level 11 Cipher Gateway Specialist. Extracts mathematically rigorous sonic fingerprints from DARKXSIDE audio archives using ffprobe / ffmpeg lavfi (ebur128 R128 loudness, aspectralstats, silencedetect). Groups tracks by sound analysis — NOT by file metadata — into named sonic constellations, writes M3U8 playlists, checkpoints into Cipher Memory, and emits geometry events on the NATS Geometry Bus for Holographic CHIT Block generation. keywords: beats, audio analysis, darkxside, cipher, ffmpeg, lavfi, ebur128, spectral, chit, holographic, m3u8, playlists, glances, nats version: 2.0.0 category: Media/Analysis/Cipher tier: 5 agent_class: Specialized agent_id: cipher_beats_analyst

PMOVES Cipher Beats Analyst

Agent Class: Specialized (Pmoves-)
Category: Media/Analysis/Cipher
Version: 2.0.0
Tier: 5 (Media) + 6 (Agent/Cipher)
Status: Level 11 — Cipher Gateway Specialist
Cipher Category: agent_checkpoint + beats_fingerprint

Mission Statement

"The beats are not music files. They are 15 years of mathematically tuned geometry— EQ'd for emotional resonance at BPM-dependent scales, compressed for holographic listening across every speaker set. This agent reads that geometry. It does not rename your files. It reveals their structure."

This specialist treats every audio file as a dimensional object:

• ebur128 — loudness integrated across time (the feeling)
• aspectralstats — spectral centroid / flatness / entropy (the color)
• Cluster distance — the relational geometry between tracks

Output: named sonic constellations → M3U8 playlists + CHIT geometry events.

Capabilities

Trigger Phrases (Pinokio 7 Interpreter — zero config)

The ffmpeg Analysis Stack (No External ML Deps)

Instead of bloated Python audio ML libraries, this agent uses ffmpeg 8.x lavfi directly. This means it runs natively alongside every PMOVES service that already depends on ffmpeg.

Audio File (.wav / .opus / .mp3 / .flac / .m4a)
  │
  ├── ffprobe → duration, bit_rate, sample_rate, channels, title tags
  │
  ├── ffmpeg -af "ebur128" → R128 Integrated Loudness (LUFS), LRA, True Peak
  │       └── [emotional weight / perceived intensity]
  │
  └── ffmpeg -af "aspectralstats,ametadata=print:file=-" (first 30s sample)
          ├── Centroid  → timbral character (bass-heavy / warm / electric / airy)
          └── Flatness  → texture (tonal / textured / noisy)

Future Upgrade Path (Phase 11 — Holographic Blocks)

When Hi-RAG multimodal Gemini embeddings go live, replace the spectral heuristics with /hirag/ingest/audio → true multi-dimensional embedding → better cluster geometry. The same M3U8 + Cipher output interface is preserved.

Sonic Group Naming Convention

Each group is named: {TempoZone}_{TimbreZone}_{EnergyZone}

Example: Andante_warm_Deep — slow, rich-toned, quiet-intensity track group.

Output Files

pmoves/data/beats/playlists/
  ├── Andante_warm_Deep.m3u8        ← sorted by BPM ascending within group
  ├── Allegro_electric_Bright.m3u8
  ├── Moderato_balanced_Mid.m3u8
  ├── ...
  └── groups_summary.json           ← machine-readable group manifest
pmoves/data/beats/soundcloud/darkxside/
  └── .fingerprints.json            ← cached ffmpeg analysis results (use --cache to reuse)

Pipeline Integration

Cipher Memory (port 8105)

Checkpoints are written at two stages:

1. fingerprint_complete — all files analyzed, before clustering
2. grouping_complete — full group manifest (resumable if interrupted)

# Query latest checkpoint
curl http://localhost:8105/api/memory/search?q=analyze_beats&category=agent_checkpoint

NATS Geometry Bus

After grouping, each constellation is published to the Geometry Bus:

Subject: pmoves.darkxside.beats.group.v1
Payload: { "group": "Andante_warm_Deep", "count": 12, "tracks": [...] }

The chit_a2ui_bridge.py consumes these to generate Holographic CHIT Blocks.

Glances Live Monitoring (port 61208)

Pushes analyze_beats.tracks_done=N events to Glances REST API during extraction so you can watch GPU/CPU usage in real-time while the analysis runs across the mesh.

# Start Glances (if not already running)
glances -w --port 61208

Full Usage Reference

# One-shot: analyze the 82 SoundCloud tracks (sample dataset)
uv run pmoves/tools/analyze_beats.py analyze \
  --input pmoves/data/beats/soundcloud/darkxside \
  --output pmoves/data/beats/playlists \
  --groups 8 \
  --glances http://localhost:61208

# Use cached fingerprints (re-cluster only — much faster)
uv run pmoves/tools/analyze_beats.py analyze --cache --groups 12

# View groups table
uv run pmoves/tools/analyze_beats.py groups

# Check Cipher + Glances health
uv run pmoves/tools/analyze_beats.py status

Environment Variables

BEATS_INPUT=pmoves/data/beats/soundcloud/darkxside
BEATS_OUTPUT=pmoves/data/beats/playlists
NATS_URL=nats://localhost:4222
CIPHER_URL=http://localhost:8105
GLANCES_URL=http://localhost:61208

Cross-Machine Invocation (Tailscale Mesh)

The analysis is CPU/IO-bound. Delegate to the 5090 node for large archives:

# From any machine on pmoves-net via SSH/Tailscale:
ssh 5090 "cd PMOVES.AI && uv run pmoves/tools/analyze_beats.py analyze \
  --input /data/beats/gdrive \
  --output /data/beats/playlists \
  --nats nats://z890:4222 \
--cipher http://z890:8105"

Agent Registry Entry

agent_id: cipher_beats_analyst
class: Specialized
tier: [5, 6]
cipher_categories: [agent_checkpoint, beats_fingerprint]
nats_subjects:
  publish: [pmoves.darkxside.beats.group.v1, media.ingest.request.v1]
  subscribe: []
tool_path: pmoves/tools/analyze_beats.py
runtime: uv run

See Also

• analyze_beats.py — the Typer agent
• chit_a2ui_bridge.py — CGP → A2UI visualization
• pmoves-holographic-blocks/SKILL.md — full beats → CHIT geometry pipeline
• AGNOTE4482.BEATS.md — DARKXSIDE audio context
• AGENT_RESILIENCE_PATTERNS.md — Cipher checkpoint protocol