Update docs for FastEmbed default and fix Qdrant dimension handling

[automationpi]

Summary

• Updated README Tier 2 section: FastEmbed default, no OpenAI key needed
• Updated indexer workflow example to drop OPENAI_API_KEY
• Added secrets breakdown (only 3 needed)
• Split embedding models into Local (default) and API (optional)
• Fixed qdrant_store.py to auto-recreate collection on dimension mismatch

Test plan

• README renders correctly
• Indexer handles model switch without manual intervention

[github-actions]

mergelore findings

Reviewed 3 past PRs · 0 relevant decisions found · confidence threshold 0.5

⚠️ Potential conflicts

Inconsistent migration path from GitHub Action to pip-based indexer · confidence: 0.80

PR #10 recently updated version references to use the GitHub Action approach (automationpi/mergelore/indexer@v0.2.1). This current PR replaces that entire pattern with a manual pip install + script approach (pip install mergelore-indexer, then mergelore-index). Users who followed PR #10's guidance will need to completely rewrite their indexer workflows, creating confusion about the recommended deployment model.

/mergelore-acknowledge · /mergelore-override [reason] · /mergelore-update-record

Automatic collection deletion on embedding model changes · confidence: 0.70

The new code in qdrant_store.py automatically deletes and recreates Qdrant collections when embedding model dimensions change. While the README mentions this behavior ("indexer automatically recreates the Qdrant collection"), this is a destructive operation that silently removes indexed data without user confirmation. Given that PR #7 emphasized data privacy and storage transparency, this automatic deletion warrants explicit user awareness and potential safeguards.

/mergelore-acknowledge · /mergelore-override [reason] · /mergelore-update-record

---

ℹ️ Related context

• Embedding model strategy shift from API-first to local-first (confidence: 0.75): PR Update version references to v0.2.1 #10 supported OpenAI's text-embedding-3-small as the default model. This PR inverts the strategy, making local FastEmbed the default and treating API models (OpenAI, Cohere) as optional extras requiring separate installation. While this improves cost and reduces API dependencies, it's a significant architectural reversal that affects the default behavior users expect from the documented tier.

---

mergelore v0.1.0 · docs