[Priority 2] Create unified VFS abstraction for all RLM resources

[marknutter]

Problem

Fragmented access patterns across RLM:

• Memory: rlm recall, rlm remember, MCP tools
• Analysis: rlm scan, rlm chunk, rlm extract
• Hooks: direct Python imports
• TUI: direct database access
• Strategies: file I/O
• Archives: exports

Each resource has custom access code. Hard to compose operations across resources.

Proposal (from AIGNE paper analysis)

Unified virtual file system (VFS) abstraction:

from rlm.vfs import mount, list_dir, read, write, search, exec

# Unified namespace
mount('/rlm/memory/', backend=SQLiteMemory)
mount('/rlm/sessions/', backend=LocalFileSystem)
mount('/rlm/strategies/', backend=LocalFileSystem)
mount('/rlm/archive/', backend=ImmutableHistory)
mount('/rlm/scratchpad/', backend=EphemeralWorkspace)

# Uniform access
results = search('/rlm/memory/', query='security')
content = read('/rlm/sessions/abc123/manifest.json')
exec('/rlm/strategies/expand_query.py', args=['Iraq war'])

Implementation

1. Design VFS API (mount, list_dir, read, write, search, exec)
2. Create backend adapters (SQLite, filesystem, exports)
3. Incrementally migrate existing code
4. Update MCP server to expose VFS operations

Impact

• Composability (pipe operations across resources)
• Extensibility (mount new backends without core changes)
• Cleaner architecture (uniform interface)
• Future-proof (easy to add vector DBs, APIs, knowledge graphs)

Effort

5-7 days (design, implementation, migration, testing)

Related

• File system abstraction from 'Everything is Context' paper (arxiv 2512.05470)
• AIGNE's AFS module