Feature Request: Local/offline mode — call AI providers directly without LiveReview backend

[tailorgunjan93]

Feature Request: Local / Offline AI Review Mode

Problem

Currently, every staged diff is sent to Hexmos's LiveReview cloud backend before
reaching an AI provider. This is a blocker for:

• Companies with NDAs or IP-sensitive codebases
• Air-gapped environments (no outbound internet)
• Developers who want zero third-party data sharing
• Users running local models via Ollama or LMStudio

Proposed Solution

Add a mode = "local" config option that makes the CLI call the AI provider
directly — bypassing LiveReview entirely:

# ~/.lrc.toml
mode     = "local"
provider = "openai_compat"
base_url = "http://localhost:11434/v1"   # Ollama
model    = "llama3"
api_key  = "ollama"                      # dummy key for local

Data Flow Comparison

Current:  diff → LiveReview cloud → AI provider → response
Local:    diff → AI provider (direct) → response

Benefits

• No diff ever leaves the machine in local mode
• Works fully offline with Ollama / LMStudio
• Removes dependency on LiveReview API availability
• Unlocks enterprise adoption (security teams love this)

Implementation Notes

• Could reuse the same connector interface, just skip the LiveReview
  ReviewSubmit() / ReviewPoll() calls
• Local mode would lose review history sync (acceptable tradeoff)
• A --local CLI flag as an override would also be useful:
  git lrc review --local