Understand the library
Inventory supported audio, retain technical and embedded evidence, and distinguish physical Files from logical Recordings.
In active development
DeepCrate.
Local-first music-library intelligence for DJs and collectors. DeepCrate is being built to understand exact copies, underlying recordings and meaningful versions—then explain what it found without taking control of the library.
The problem
A music collection is not a folder full of interchangeable files.
Years of downloads, migrations and DJ software leave collections full of duplicate encodes, inconsistent metadata, poor filenames, edits, remixes, radio versions and extended versions. Simple duplicate removal is dangerous because similar-looking files can carry musically important differences.
The collection becomes harder to understand at exactly the point it becomes most valuable.
The insight
A file is not the same thing as a recording.
Two different Files may represent the same Recording: an MP3 and an AIFF, for example. Two similarly named Files may represent distinct versions that a DJ deliberately owns.
DeepCrate models that distinction directly. Exact byte identity, metadata, duration, filenames, acoustic evidence and version conflicts contribute evidence. None is treated as magic truth.
The product direction
Intelligence you can understand.
DeepCrate is moving from a verified engineering precursor toward a read-only desktop product. The first product milestone will not modify source music.
See relationships
Identify exact duplicates, probable same recordings, alternate representations and evidence that remains uncertain.
Protect versions
Make conflicting Radio Edit, Extended Mix, remix and other version evidence a reason for caution—not automatic collapse.
Explain the conclusion
Show the outcome first, plain reasons next and technical provenance only when the user wants it.
Safety by design
Confidence is not permission.
A strong analytical conclusion is still not authority to change a music collection. DeepCrate separates observation, evidence, recommendation, user judgement and action permission.
The engineering precursor already demonstrates frozen plans, exact gates, rehearsal, journalling, verification, interruption handling and rollback under controlled tests. The first consumer product deliberately stays read-only while packaging, broader failure evidence and user comprehension catch up.
Analyse before acting. Preserve uncertainty when it matters.
Local first
The collection stays under the DJ's control.
Inventory, hashes, fingerprints, identity rules and catalogue queries naturally fit local execution. Core use should not require uploading a large copyrighted library, maintaining developer API keys or depending on an AI account.
Optional online evidence may add value later, but it must remain explicit, attributed and replaceable.
How we built it
The product process is part of the evidence.
DeepCrate demonstrates how Cobalt Giraffe moves from an awkward real problem to an executable product definition.
- 01Problem
- 02Users
- 03Requirements
- 04Architecture
- 05Safety
- 06Engineering
- 07Evidence
Current status
Substantial foundations. Not commercially available yet.
- Implemented precursor
- Recursive and incremental scanning, exact hashes, recording-identity foundations, version protection, provenance, SQLite persistence and controlled safety machinery.
- Recorded scale point
- One approximately 130 GB / 3,650-file DJ archive processed successfully. This is feasibility evidence, not a universal performance promise.
- Next product milestone
- An installable Windows-first, read-only library intelligence product with review, ownership search, durable decisions and explicit evidence.
- Availability
- In active development. No launch date or commercial availability is being claimed.