feat: configurable LLM features + thresholds for memory

MemoryConfig gains:
- llm_search (default true): LLM-based episode ranking
- llm_extract (default true): LLM-based fact extraction
- llm_consolidate (default true): LLM-based fact consolidation
- merge_threshold (default 0.7): go-vector cosine for auto-merge
- add_threshold (default 0.3): go-vector cosine for auto-add

MergeDetector supports per-instance thresholds via
NewMergeDetectorWithThresholds().

Author: jkyberneees

docs/CONFIG.md

@@ -123,7 +123,12 @@ The `memory` section controls the persistent memory system (see [docs/MEMORY.md]
     "buffer_lines": 20,
     "buffer_enabled": true,
     "merge_on_write": true,
-    "extract_on_end": true
+    "extract_on_end": true,
+    "llm_search": true,
+    "llm_extract": true,
+    "llm_consolidate": true,
+    "merge_threshold": 0.7,
+    "add_threshold": 0.3
   }
 }
 ```
@@ -137,6 +142,11 @@ The `memory` section controls the persistent memory system (see [docs/MEMORY.md]
 | `buffer_enabled` | true | Enable the turn-level buffer |
 | `merge_on_write` | true | Use go-vector RP similarity to auto-merge related entries |
 | `extract_on_end` | true | Extract durable facts via LLM at session end (≥3 turns) |
+| `llm_search` | true | Use LLM to rank episode search results by relevance |
+| `llm_extract` | true | Use LLM for end-of-session fact extraction |
+| `llm_consolidate` | true | Use LLM to merge related fact entries |
+| `merge_threshold` | 0.7 | go-vector cosine threshold for auto-merge (0.0–1.0) |
+| `add_threshold` | 0.3 | go-vector cosine threshold for auto-add (0.0–1.0) |
 
 ## Sub-agent configuration
 

internal/memory/memory.go

@@ -15,13 +15,18 @@ const defaultBufferLines = 20
 // MemoryConfig holds configuration for the memory system.
 // Mirrors the JSON config section.
 type MemoryConfig struct {
-	Enabled        bool `json:"enabled"`
-	FactsLimitUser int  `json:"facts_limit_user"`
-	FactsLimitEnv  int  `json:"facts_limit_env"`
-	BufferLines    int  `json:"buffer_lines"`
-	BufferEnabled  bool `json:"buffer_enabled"`
-	MergeOnWrite   bool `json:"merge_on_write"`
-	ExtractOnEnd   bool `json:"extract_on_end"`
+	Enabled        bool    `json:"enabled"`
+	FactsLimitUser int     `json:"facts_limit_user"`
+	FactsLimitEnv  int     `json:"facts_limit_env"`
+	BufferLines    int     `json:"buffer_lines"`
+	BufferEnabled  bool    `json:"buffer_enabled"`
+	MergeOnWrite   bool    `json:"merge_on_write"`
+	ExtractOnEnd   bool    `json:"extract_on_end"`
+	LLMSearch      bool    `json:"llm_search"`
+	LLMExtract     bool    `json:"llm_extract"`
+	LLMConsolidate bool    `json:"llm_consolidate"`
+	MergeThreshold float32 `json:"merge_threshold"`
+	AddThreshold   float32 `json:"add_threshold"`
 }
 
 // DefaultMemoryConfig returns sensible defaults.
@@ -34,6 +39,11 @@ func DefaultMemoryConfig() MemoryConfig {
 		BufferEnabled:  true,
 		MergeOnWrite:   true,
 		ExtractOnEnd:   true,
+		LLMSearch:      true,
+		LLMExtract:     true,
+		LLMConsolidate: true,
+		MergeThreshold: MergeThreshold,
+		AddThreshold:   AddThreshold,
 	}
 }
 
@@ -68,18 +78,24 @@ func NewMemoryManager(memoryDir string, llc LLMClient, cfg MemoryConfig) *Memory
 	if cfg.BufferLines <= 0 {
 		cfg.BufferLines = defaultBufferLines
 	}
+	if cfg.MergeThreshold <= 0 {
+		cfg.MergeThreshold = MergeThreshold
+	}
+	if cfg.AddThreshold <= 0 {
+		cfg.AddThreshold = AddThreshold
+	}
 
 	factsDir := memoryDir
 	episodesDir := memoryDir
 
 	factStore := NewFactStore(factsDir, cfg.FactsLimitUser, cfg.FactsLimitEnv)
-	// Use LLM-based episode ranker when an LLM client is available
+	// Use LLM-based episode ranker when an LLM client is available and enabled
 	var rankFn RankStrategy
-	if llc != nil {
+	if llc != nil && cfg.LLMSearch {
 		rankFn = NewLLMRanker(llc)
 	}
 	episodeStore := NewEpisodeStore(episodesDir, rankFn)
-	mergeDetector := NewMergeDetector(0) // default 256 dims
+	mergeDetector := NewMergeDetectorWithThresholds(0, cfg.MergeThreshold, cfg.AddThreshold)
 
 	return &MemoryManager{
 		facts:    factStore,
@@ -206,12 +222,13 @@ func (m *MemoryManager) ReadFacts() (userContent, envContent string, err error)
 }
 
 // Consolidate uses the LLM to merge related entries in a target file
-// for better density. Falls back to no-op if no LLM is available.
+// for better density. Falls back to no-op if LLM is unavailable or
+// LLMConsolidate is disabled in config.
 func (m *MemoryManager) Consolidate(target string) error {
 	if !m.cfg.Enabled {
 		return fmt.Errorf("memory: disabled")
 	}
-	if m.llm == nil {
+	if m.llm == nil || !m.cfg.LLMConsolidate {
 		return fmt.Errorf("memory: consolidation requires LLM client")
 	}
 
@@ -288,7 +305,7 @@ func (m *MemoryManager) ClearBuffer() {
 // OnSessionEnd is called when a session ends. If turns >= threshold,
 // extracts durable facts using the LLM and stores them as an episode.
 func (m *MemoryManager) OnSessionEnd(sessionID string, turns int, messages []string) {
-	if !m.cfg.ExtractOnEnd || m.llm == nil || turns < 3 || len(messages) == 0 {
+	if !m.cfg.ExtractOnEnd || !m.cfg.LLMExtract || m.llm == nil || turns < 3 || len(messages) == 0 {
 		return
 	}
 

internal/memory/merge.go

@@ -29,22 +29,43 @@ const (
 //  2. Fit(corpus) — builds vocabulary from existing entries
 //  3. Classify(entry) → action + similarIdx + similarity
 //  4. After facts change → Fit(newCorpus) to rebuild vocabulary
+//
+// Thresholds control the classification:
+//   - mergeThreshold: cosine above this → auto-merge (default 0.7)
+//   - addThreshold: cosine below this → auto-add (default 0.3)
+//   - Between thresholds: "judge" — requires LLM to decide
 type MergeDetector struct {
-	rp     *vector.RandomProjections
-	corpus []string
-	vecs   []vector.Vector // precomputed embeddings of corpus
-	dims   int
+	rp             *vector.RandomProjections
+	corpus         []string
+	vecs           []vector.Vector // precomputed embeddings of corpus
+	dims           int
+	mergeThreshold float32
+	addThreshold   float32
 }
 
 // NewMergeDetector creates a MergeDetector with the given output
 // dimensionality for the RP embedder. Pass 0 for default (256).
+// Uses default thresholds (0.7 merge, 0.3 add).
 func NewMergeDetector(dims int) *MergeDetector {
+	return NewMergeDetectorWithThresholds(dims, MergeThreshold, AddThreshold)
+}
+
+// NewMergeDetectorWithThresholds creates a MergeDetector with custom thresholds.
+func NewMergeDetectorWithThresholds(dims int, mergeThreshold, addThreshold float32) *MergeDetector {
 	if dims <= 0 {
 		dims = defaultOutputDim
 	}
+	if mergeThreshold <= 0 {
+		mergeThreshold = MergeThreshold
+	}
+	if addThreshold <= 0 || addThreshold >= mergeThreshold {
+		addThreshold = AddThreshold
+	}
 	return &MergeDetector{
-		rp:   vector.NewRandomProjections(dims),
-		dims: dims,
+		rp:             vector.NewRandomProjections(dims),
+		dims:           dims,
+		mergeThreshold: mergeThreshold,
+		addThreshold:   addThreshold,
 	}
 }
 
@@ -111,9 +132,9 @@ func (m *MergeDetector) Classify(entry string) (action string, similarIdx int, s
 	similarIdx = bestIdx
 
 	switch {
-	case bestSim >= MergeThreshold:
+	case bestSim >= m.mergeThreshold:
 		return "merge", bestIdx, bestSim
-	case bestSim <= AddThreshold:
+	case bestSim <= m.addThreshold:
 		return "add", bestIdx, bestSim
 	default:
 		return "judge", bestIdx, bestSim