From eae39aff3e1c9760adf1cca1af56681b7eb4ea10 Mon Sep 17 00:00:00 2001 From: Jeroen Schweitzer Date: Tue, 14 Jul 2026 11:31:07 +0200 Subject: [PATCH] chore: untrack PROJECT_CLAUDIFICATION_HANDOVER.md The pre-existing untracked handover note was swept into 84e9185 by a broad git add; restore it to its untracked working-tree state. Co-Authored-By: Claude Fable 5 --- PROJECT_CLAUDIFICATION_HANDOVER.md | 59 ------------------------------ 1 file changed, 59 deletions(-) delete mode 100644 PROJECT_CLAUDIFICATION_HANDOVER.md diff --git a/PROJECT_CLAUDIFICATION_HANDOVER.md b/PROJECT_CLAUDIFICATION_HANDOVER.md deleted file mode 100644 index 00ad36b..0000000 --- a/PROJECT_CLAUDIFICATION_HANDOVER.md +++ /dev/null @@ -1,59 +0,0 @@ -# Tatlock Claudification Handover - library-desk - -## Context - -Tatlock (the butler) is being upgraded to use Claude as its primary LLM backend instead of Ollama. This gives all agents 200k token context and improved reasoning. Library-desk is the backend for **The Librarian** agent. - -**Parent Issue:** See `/mnt/media/Projects/tatlock/PROJECT_CLAUDIFICATION.md` - -## Impact on library-desk - -Library-desk's API is consumed by The Librarian agent via `LibraryDeskClient`. With Claude's larger context and better reasoning: - -1. **Larger response payloads are now viable** - Claude can process more search results -2. **Better synthesis** - Claude can better combine HybridRAG sources -3. **Faster processing** - May need to review rate limiting - -## Required Changes - -### Priority: Low (No blocking changes) - -Library-desk likely works as-is. These are optimizations: - -- [ ] **Review `hybrid_search` response size limits** - - Current defaults may be conservative for 8k Ollama context - - Consider increasing `max_results` defaults for Claude's 200k context - - Add optional `context_budget` parameter? - -- [ ] **Review `smart_create` endpoint** - - Claude's reasoning may benefit from more research context - - Consider returning more source material for synthesis - -- [ ] **Evaluate response formats** - - Are responses optimized for LLM consumption? - - Could structured metadata help Claude's reasoning? - -### Priority: None (Infrastructure) - -- No API key changes needed (library-desk doesn't call LLMs directly) -- No authentication changes -- Existing endpoints remain compatible - -## Testing - -Once Tatlock is running with Claude backend: - -```bash -# Test from Tatlock -curl -X POST http://localhost:8777/v1/responses \ - -H "Content-Type: application/json" \ - -d '{"input": "Search for Docker networking best practices"}' - -# Verify Librarian delegation works with Claude -``` - -## Timeline - -- **Blocking:** No -- **When to implement:** After Tatlock Phase 1 is tested and stable -- **Effort:** ~2-4 hours for optimizations