refactor(config): split threshold into search_threshold and ask_threshold

Separate config fields so search and ask can be tuned independently.
Both default to 0.001. Bump to v1.0.7.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: ariel-frischer

README.md

@@ -125,7 +125,8 @@ sources = [
 # embed_dims = 1536
 # chat_model = "gpt-4o-mini"
 # max_chunk_chars = 2000
-# threshold = 0.001             # minimum cosine similarity to include a result (also --threshold flag)
+# search_threshold = 0.001      # min cosine similarity for `kb search` (also --threshold flag)
+# ask_threshold = 0.001         # min cosine similarity for `kb ask` (also --threshold flag)
 # rerank_fetch_k = 20
 # rerank_top_k = 5
 # index_code = false       # set true to also index source code files

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "kb"
-version = "1.0.6"
+version = "1.0.7"
 description = "CLI knowledge base: index markdown + PDFs, hybrid search, RAG answers. Powered by sqlite-vec."
 readme = "README.md"
 license = "MIT"

src/kb/cli.py

@@ -42,7 +42,7 @@
   kb sources                     List configured sources
   kb index [DIR...] [--no-size-limit]  Index sources (skip files > max_file_size_mb)
   kb allow <file>                Whitelist a large file for indexing
-  kb search "query" [k]          Hybrid semantic + keyword search (default k=5)
+  kb search "query" [k] [--threshold N]  Hybrid semantic + keyword search (default k=5, threshold=0.001)
   kb ask "question" [k] [--threshold N]  RAG: search + rerank + answer (default k=8, threshold=0.001)
   kb similar <file> [k]          Find similar documents (no API call, default k=10)
   kb tag <file> tag1 [tag2...]   Add tags to a document
@@ -219,7 +219,7 @@ def cmd_index(cfg: Config, args: list[str]):
 
 def cmd_search(query: str, cfg: Config, top_k: int = 5, threshold: float | None = None):
     if threshold is not None:
-        cfg.threshold = threshold
+        cfg.search_threshold = threshold
     if not cfg.db_path.exists():
         print("No index found. Run 'kb index' first.")
         sys.exit(1)
@@ -240,7 +240,7 @@ def cmd_search(query: str, cfg: Config, top_k: int = 5, threshold: float | None
     query_emb = resp.data[0].embedding
     embed_ms = (time.time() - t0) * 1000
 
-    has_threshold = cfg.threshold > 0
+    has_threshold = cfg.search_threshold > 0
     retrieve_k = (top_k * 5) if has_filters else (top_k * 3)
 
     t0 = time.time()
@@ -282,7 +282,7 @@ def cmd_search(query: str, cfg: Config, top_k: int = 5, threshold: float | None
         results = [
             r
             for r in results
-            if r["similarity"] is None or r["similarity"] >= cfg.threshold
+            if r["similarity"] is None or r["similarity"] >= cfg.search_threshold
         ]
 
     print(f'Query: "{clean_query}"')
@@ -319,7 +319,7 @@ def cmd_search(query: str, cfg: Config, top_k: int = 5, threshold: float | None
 def cmd_ask(question: str, cfg: Config, top_k: int = 8, threshold: float | None = None):
     """Full RAG: hybrid retrieve -> filter -> LLM rerank -> confidence filter -> answer."""
     if threshold is not None:
-        cfg.threshold = threshold
+        cfg.ask_threshold = threshold
     if not cfg.db_path.exists():
         print("No index found. Run 'kb index' first.")
         sys.exit(1)
@@ -380,7 +380,7 @@ def cmd_ask(question: str, cfg: Config, top_k: int = 8, threshold: float | None
     filtered = [
         r
         for r in results
-        if r["similarity"] is None or r["similarity"] >= cfg.threshold
+        if r["similarity"] is None or r["similarity"] >= cfg.ask_threshold
     ]
 
     if not filtered:

src/kb/config.py

@@ -39,10 +39,11 @@
 # min_chunk_chars = 50
 
 # Search
-# threshold = 0.001        # minimum cosine similarity to include a result (0.0–1.0)
-# rrf_k = 60.0            # RRF smoothing constant
-# rerank_fetch_k = 20     # candidates to fetch for LLM rerank
-# rerank_top_k = 5        # how many to keep after rerank
+# search_threshold = 0.001  # min cosine similarity for `kb search` (0.0–1.0)
+# ask_threshold = 0.001     # min cosine similarity for `kb ask` (0.0–1.0)
+# rrf_k = 60.0              # RRF smoothing constant
+# rerank_fetch_k = 20       # candidates to fetch for LLM rerank
+# rerank_top_k = 5          # how many to keep after rerank
 
 # Format options
 # index_code = false                # also index source code files (.py, .js, .ts, etc.)
@@ -91,7 +92,8 @@ class Config:
     chat_model: str = "gpt-4o-mini"
     max_chunk_chars: int = 2000
     min_chunk_chars: int = 50
-    threshold: float = 0.001
+    search_threshold: float = 0.001
+    ask_threshold: float = 0.001
     rrf_k: float = 60.0
     rerank_fetch_k: int = 20
     rerank_top_k: int = 5

tests/test_cli_commands.py

@@ -186,7 +186,7 @@ def test_search_top_k(self, populated_db, capsys):
 
     def test_threshold_reduces_result_count(self, tmp_path, capsys):
         """Threshold should remove low-similarity results, not backfill with FTS-only."""
-        cfg = Config(embed_dims=4, threshold=0.99)
+        cfg = Config(embed_dims=4, search_threshold=0.99)
         cfg.scope = "project"
         cfg.config_dir = tmp_path
         cfg.config_path = tmp_path / ".kb.toml"
@@ -300,7 +300,7 @@ def test_ask_calls_rerank_when_enough_results(self, populated_db, capsys):
     def test_ask_no_results_above_threshold(self, tmp_path, capsys):
         """When all results have similarity below threshold, show 'no relevant documents'."""
         # Build a DB where vec results have high distance (low similarity)
-        cfg = Config(embed_dims=4, threshold=0.99)
+        cfg = Config(embed_dims=4, ask_threshold=0.99)
         cfg.scope = "project"
         cfg.config_dir = tmp_path
         cfg.config_path = tmp_path / ".kb.toml"

tests/test_config.py

@@ -22,7 +22,8 @@ def test_defaults(self):
         assert cfg.chat_model == "gpt-4o-mini"
         assert cfg.max_chunk_chars == 2000
         assert cfg.min_chunk_chars == 50
-        assert cfg.threshold == 0.001
+        assert cfg.search_threshold == 0.001
+        assert cfg.ask_threshold == 0.001
         assert cfg.rrf_k == 60.0
         assert cfg.rerank_fetch_k == 20
         assert cfg.rerank_top_k == 5