mirror of
https://github.com/samvallad33/vestige.git
synced 2026-05-09 07:42:37 +02:00
feat: v2.0.4 "Deep Reference" — cognitive reasoning engine + 10 bug fixes
New features: - deep_reference tool (#22): 8-stage cognitive reasoning pipeline with FSRS-6 trust scoring, intent classification (FactCheck/Timeline/RootCause/Comparison/ Synthesis), spreading activation expansion, temporal supersession, trust-weighted contradiction analysis, relation assessment, dream insight integration, and algorithmic reasoning chain generation — all without calling an LLM - cross_reference (#23): backward-compatible alias for deep_reference - retrieval_mode parameter on search (precise/balanced/exhaustive) - get_batch action on memory tool (up to 20 IDs per call) - Token budget raised from 10K to 100K on search + session_context - Dates (createdAt/updatedAt) on all search results and session_context lines Bug fixes (GitHub Issue #25 — all 10 resolved): - state_transitions empty: wired record_memory_access into strengthen_batch - chain/bridges no storage fallback: added with edge deduplication - knowledge_edges dead schema: documented as deprecated - insights not persisted from dream: wired save_insight after generation - find_duplicates threshold dropped: serde alias fix - search min_retention ignored: serde aliases for snake_case params - intention time triggers null: removed dead trigger_at embedding - changelog missing dreams: added get_dream_history + event integration - phantom Related IDs: clarified message text - fsrs_cards empty: documented as harmless dead schema Security hardening: - HTTP transport CORS: permissive() → localhost-only - Auth token panic guard: &token[..8] → safe min(8) slice - UTF-8 boundary fix: floor_char_boundary on content truncation - All unwrap() removed from HTTP transport (unwrap_or_else fallback) - Dream memory_count capped at 500 (prevents O(N²) hang) - Dormant state threshold aligned (0.3 → 0.4) Stats: 23 tools, 758 tests, 0 failures, 0 warnings, 0 unwraps in production Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Sam Valladares
parent
61091e06b9
commit
04781a95e2
28 changed files with 1797 additions and 102 deletions
|
|
@ -68,9 +68,15 @@ pub fn schema() -> Value {
|
|||
},
|
||||
"token_budget": {
|
||||
"type": "integer",
|
||||
"description": "Max tokens for response. Server truncates content to fit budget. Use memory(action='get') for full content of specific IDs.",
|
||||
"description": "Max tokens for response. Server truncates content to fit budget. Use memory(action='get') for full content of specific IDs. With 1M context models, budgets up to 100K are practical.",
|
||||
"minimum": 100,
|
||||
"maximum": 10000
|
||||
"maximum": 100000
|
||||
},
|
||||
"retrieval_mode": {
|
||||
"type": "string",
|
||||
"description": "precise: top results only (fast, token-efficient, skips activation/competition). balanced: full 7-stage cognitive pipeline (default). exhaustive: maximum recall with 5x overfetch, deep graph traversal, no competition suppression.",
|
||||
"enum": ["precise", "balanced", "exhaustive"],
|
||||
"default": "balanced"
|
||||
}
|
||||
},
|
||||
"required": ["query"]
|
||||
|
|
@ -82,13 +88,18 @@ pub fn schema() -> Value {
|
|||
struct SearchArgs {
|
||||
query: String,
|
||||
limit: Option<i32>,
|
||||
#[serde(alias = "min_retention")]
|
||||
min_retention: Option<f64>,
|
||||
#[serde(alias = "min_similarity")]
|
||||
min_similarity: Option<f32>,
|
||||
#[serde(alias = "detail_level")]
|
||||
detail_level: Option<String>,
|
||||
#[serde(alias = "context_topics")]
|
||||
context_topics: Option<Vec<String>>,
|
||||
#[serde(alias = "token_budget")]
|
||||
token_budget: Option<i32>,
|
||||
#[serde(alias = "retrieval_mode")]
|
||||
retrieval_mode: Option<String>,
|
||||
}
|
||||
|
||||
/// Execute unified search with 7-stage cognitive pipeline.
|
||||
|
|
@ -135,14 +146,32 @@ pub async fn execute(
|
|||
let min_retention = args.min_retention.unwrap_or(0.0).clamp(0.0, 1.0);
|
||||
let min_similarity = args.min_similarity.unwrap_or(0.5).clamp(0.0, 1.0);
|
||||
|
||||
// Validate retrieval_mode
|
||||
let retrieval_mode = match args.retrieval_mode.as_deref() {
|
||||
Some("precise") => "precise",
|
||||
Some("exhaustive") => "exhaustive",
|
||||
Some("balanced") | None => "balanced",
|
||||
Some(invalid) => {
|
||||
return Err(format!(
|
||||
"Invalid retrieval_mode '{}'. Must be 'precise', 'balanced', or 'exhaustive'.",
|
||||
invalid
|
||||
));
|
||||
}
|
||||
};
|
||||
|
||||
// Favor semantic search — research shows 0.3/0.7 outperforms equal weights
|
||||
let keyword_weight = 0.3_f32;
|
||||
let semantic_weight = 0.7_f32;
|
||||
|
||||
// ====================================================================
|
||||
// STAGE 1: Hybrid search with 3x over-fetch for reranking pool
|
||||
// STAGE 1: Hybrid search with Nx over-fetch for reranking pool
|
||||
// ====================================================================
|
||||
let overfetch_limit = (limit * 3).min(100); // Cap at 100 to avoid excessive DB load
|
||||
let overfetch_multiplier = match retrieval_mode {
|
||||
"precise" => 1, // No overfetch — return exactly what's asked
|
||||
"exhaustive" => 5, // Deep overfetch for maximum recall
|
||||
_ => 3, // Balanced default
|
||||
};
|
||||
let overfetch_limit = (limit * overfetch_multiplier).min(100); // Cap at 100 to avoid excessive DB load
|
||||
|
||||
let results = storage
|
||||
.hybrid_search(&args.query, overfetch_limit, keyword_weight, semantic_weight)
|
||||
|
|
@ -215,7 +244,7 @@ pub async fn execute(
|
|||
// Determine state from retention strength
|
||||
lifecycle.state = if result.node.retention_strength > 0.7 {
|
||||
MemoryState::Active
|
||||
} else if result.node.retention_strength > 0.3 {
|
||||
} else if result.node.retention_strength > 0.4 {
|
||||
MemoryState::Dormant
|
||||
} else if result.node.retention_strength > 0.1 {
|
||||
MemoryState::Silent
|
||||
|
|
@ -275,9 +304,11 @@ pub async fn execute(
|
|||
|
||||
// ====================================================================
|
||||
// STAGE 5B: Retrieval competition (Anderson et al. 1994)
|
||||
// Skipped in precise mode (no need) and exhaustive mode (want all results)
|
||||
// ====================================================================
|
||||
let mut suppressed_count = 0_usize;
|
||||
if filtered_results.len() > 1
|
||||
if retrieval_mode == "balanced"
|
||||
&& filtered_results.len() > 1
|
||||
&& let Ok(mut cog) = cognitive.try_lock()
|
||||
{
|
||||
let candidates: Vec<CompetitionCandidate> = filtered_results
|
||||
|
|
@ -321,21 +352,31 @@ pub async fn execute(
|
|||
|
||||
// ====================================================================
|
||||
// STAGE 6: Spreading activation (find associated memories)
|
||||
// Skipped in precise mode. Deeper (5 results) in exhaustive mode.
|
||||
// ====================================================================
|
||||
let associations: Vec<Value> = if let Ok(mut cog) = cognitive.try_lock() {
|
||||
if let Some(first) = filtered_results.first() {
|
||||
let activated = cog.activation_network.activate(&first.node.id, 1.0);
|
||||
activated
|
||||
.iter()
|
||||
.take(3)
|
||||
.map(|a| {
|
||||
serde_json::json!({
|
||||
"memoryId": a.memory_id,
|
||||
"activation": a.activation,
|
||||
"distance": a.distance,
|
||||
let activation_take = match retrieval_mode {
|
||||
"precise" => 0, // Skip entirely
|
||||
"exhaustive" => 5, // Deeper graph traversal
|
||||
_ => 3, // Balanced default
|
||||
};
|
||||
let associations: Vec<Value> = if activation_take > 0 {
|
||||
if let Ok(mut cog) = cognitive.try_lock() {
|
||||
if let Some(first) = filtered_results.first() {
|
||||
let activated = cog.activation_network.activate(&first.node.id, 1.0);
|
||||
activated
|
||||
.iter()
|
||||
.take(activation_take)
|
||||
.map(|a| {
|
||||
serde_json::json!({
|
||||
"memoryId": a.memory_id,
|
||||
"activation": a.activation,
|
||||
"distance": a.distance,
|
||||
})
|
||||
})
|
||||
})
|
||||
.collect()
|
||||
.collect()
|
||||
} else {
|
||||
vec![]
|
||||
}
|
||||
} else {
|
||||
vec![]
|
||||
}
|
||||
|
|
@ -401,7 +442,7 @@ pub async fn execute(
|
|||
let mut budget_expandable: Vec<String> = Vec::new();
|
||||
let mut budget_tokens_used: Option<usize> = None;
|
||||
if let Some(budget) = args.token_budget {
|
||||
let budget = budget.clamp(100, 10000) as usize;
|
||||
let budget = budget.clamp(100, 100000) as usize;
|
||||
let budget_chars = budget * 4;
|
||||
let mut used = 0;
|
||||
let mut budgeted = Vec::new();
|
||||
|
|
@ -428,11 +469,17 @@ pub async fn execute(
|
|||
let mut response = serde_json::json!({
|
||||
"query": args.query,
|
||||
"method": "hybrid+cognitive",
|
||||
"retrievalMode": retrieval_mode,
|
||||
"detailLevel": detail_level,
|
||||
"total": formatted.len(),
|
||||
"results": formatted,
|
||||
});
|
||||
|
||||
// Helpful hint when no results found
|
||||
if formatted.is_empty() {
|
||||
response["hint"] = serde_json::json!("No memories found. Use smart_ingest to add memories, or try a broader query.");
|
||||
}
|
||||
|
||||
// Include associations if any were found
|
||||
if !associations.is_empty() {
|
||||
response["associations"] = serde_json::json!(associations);
|
||||
|
|
@ -499,7 +546,7 @@ fn format_search_result(r: &vestige_core::SearchResult, detail_level: &str) -> V
|
|||
"validUntil": r.node.valid_until.map(|dt| dt.to_rfc3339()),
|
||||
"matchType": format!("{:?}", r.match_type),
|
||||
}),
|
||||
// "summary" (default) — backwards compatible
|
||||
// "summary" (default) — includes dates so AI never has to guess when a memory is from
|
||||
_ => serde_json::json!({
|
||||
"id": r.node.id,
|
||||
"content": r.node.content,
|
||||
|
|
@ -509,6 +556,8 @@ fn format_search_result(r: &vestige_core::SearchResult, detail_level: &str) -> V
|
|||
"nodeType": r.node.node_type,
|
||||
"tags": r.node.tags,
|
||||
"retentionStrength": r.node.retention_strength,
|
||||
"createdAt": r.node.created_at.to_rfc3339(),
|
||||
"updatedAt": r.node.updated_at.to_rfc3339(),
|
||||
}),
|
||||
}
|
||||
}
|
||||
|
|
@ -1004,10 +1053,11 @@ mod tests {
|
|||
let results = value["results"].as_array().unwrap();
|
||||
if !results.is_empty() {
|
||||
let first = &results[0];
|
||||
// Summary should have content but not timestamps
|
||||
// Summary should have content AND timestamps (v2.1: dates always visible)
|
||||
assert!(first["content"].is_string());
|
||||
assert!(first["id"].is_string());
|
||||
assert!(first.get("createdAt").is_none() || first["createdAt"].is_null());
|
||||
assert!(first["createdAt"].is_string(), "summary must include createdAt");
|
||||
assert!(first["updatedAt"].is_string(), "summary must include updatedAt");
|
||||
}
|
||||
}
|
||||
|
||||
|
|
@ -1106,6 +1156,6 @@ mod tests {
|
|||
let tb = &schema_value["properties"]["token_budget"];
|
||||
assert!(tb.is_object());
|
||||
assert_eq!(tb["minimum"], 100);
|
||||
assert_eq!(tb["maximum"], 10000);
|
||||
assert_eq!(tb["maximum"], 100000);
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue