v1.0.0: Intelligence — candle runtime, orchestrator, reranker, query expansion (#10)

Runtime migration from ONNX to candle (pure Rust ML). Adds LLM-powered search intelligence: research orchestrator with query expansion, cross-encoder reranking as 4th RRF lane, and adaptive lane weights per query intent. Intelligence is opt-in via engraph configure --enable-intelligence.

18 tasks, 271 tests, +3675/-945 lines across 18 files.

Author: devwhodevs

Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "engraph"
-version = "0.7.0"
+version = "1.0.0"
 edition = "2024"
 description = "Local knowledge graph for AI agents. Hybrid search + MCP server for Obsidian vaults."
 license = "MIT"
@@ -20,12 +20,10 @@ anyhow = "1"
 rusqlite = { version = "0.32", features = ["bundled"] }
 tracing = "0.1"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }
-ort = { version = "2.0.0-rc.12", features = ["ndarray"] }
 tokenizers = { version = "0.22", default-features = false, features = ["fancy-regex"] }
 sha2 = "0.10"
 ureq = "2.12"
 indicatif = "0.17"
-ndarray = "0.17"
 sqlite-vec = "0.1.8-alpha.1"
 zerocopy = { version = "0.7", features = ["derive"] }
 rayon = "1"
@@ -36,6 +34,14 @@ rmcp = { version = "1.2", features = ["transport-io"] }
 tokio = { version = "1", features = ["macros", "rt-multi-thread"] }
 notify = "7.0"
 notify-debouncer-full = "0.4"
+candle-core = "0.9"
+candle-nn = "0.9"
+candle-transformers = "0.9"
+
+[features]
+default = []
+metal = ["candle-core/metal"]
+cuda = ["candle-core/cuda"]
 
 [dev-dependencies]
 tempfile = "3"

Cargo.toml

@@ -0,0 +1,42 @@
+Output assets/demo-search.gif
+Set Shell zsh
+Set FontSize 14
+Set Width 1000
+Set Height 600
+Set Padding 20
+Set Theme "Catppuccin Mocha"
+Set TypingSpeed 40ms
+
+Type "# Index an Obsidian vault"
+Enter
+Sleep 500ms
+
+Type "engraph index ~/vault"
+Enter
+Sleep 3s
+
+Type ""
+Enter
+Sleep 300ms
+
+Type "# Search across notes — 3-lane hybrid (semantic + keyword + graph)"
+Enter
+Sleep 500ms
+
+Type "engraph search 'how does authentication work' --explain"
+Enter
+Sleep 4s
+
+Type ""
+Enter
+Sleep 300ms
+
+Type "# Get rich context for an AI agent"
+Enter
+Sleep 500ms
+
+Type "engraph context who 'Steve Barbera'"
+Enter
+Sleep 3s
+
+Sleep 2s

assets/demo-search.tape

@@ -1,9 +1,21 @@
 use anyhow::{Context, Result};
-use serde::Deserialize;
-use std::path::PathBuf;
+use serde::{Deserialize, Serialize};
+use std::path::{Path, PathBuf};
+
+/// Model override configuration.
+#[derive(Debug, Clone, Default, Serialize, Deserialize)]
+#[serde(default)]
+pub struct ModelConfig {
+    /// Override embedding model URI (e.g., "hf:repo/file.gguf").
+    pub embed: Option<String>,
+    /// Override reranker model URI.
+    pub rerank: Option<String>,
+    /// Override expansion/orchestrator model URI.
+    pub expand: Option<String>,
+}
 
 /// Application configuration, loaded from `~/.engraph/config.toml` with CLI overrides.
-#[derive(Debug, Clone, Deserialize)]
+#[derive(Debug, Clone, Serialize, Deserialize)]
 #[serde(default)]
 pub struct Config {
     /// Path to the Obsidian vault to index.
@@ -14,6 +26,10 @@ pub struct Config {
     pub exclude: Vec<String>,
     /// Number of files to process per embedding batch.
     pub batch_size: usize,
+    /// Whether intelligence features are enabled. None = not yet configured.
+    pub intelligence: Option<bool>,
+    /// Model override URIs.
+    pub models: ModelConfig,
 }
 
 impl Default for Config {
@@ -23,6 +39,8 @@ impl Default for Config {
             top_n: 5,
             exclude: vec![".obsidian/".to_string()],
             batch_size: 64,
+            intelligence: None,
+            models: ModelConfig::default(),
         }
     }
 }
@@ -68,6 +86,34 @@ impl Config {
         let dir = Self::data_dir()?;
         crate::profile::load_vault_toml(&dir)
     }
+
+    /// Whether intelligence is enabled (defaults to false if not configured).
+    pub fn intelligence_enabled(&self) -> bool {
+        self.intelligence.unwrap_or(false)
+    }
+
+    /// Save config to a specific path.
+    pub fn save_to(&self, path: &Path) -> Result<()> {
+        let content = toml::to_string_pretty(self).context("serializing config")?;
+        std::fs::write(path, content).with_context(|| format!("writing {}", path.display()))?;
+        Ok(())
+    }
+
+    /// Load config from a specific path.
+    pub fn load_from(path: &Path) -> Result<Self> {
+        let contents =
+            std::fs::read_to_string(path).with_context(|| format!("reading {}", path.display()))?;
+        let config: Config =
+            toml::from_str(&contents).with_context(|| format!("parsing {}", path.display()))?;
+        Ok(config)
+    }
+
+    /// Save to the default config path (`~/.engraph/config.toml`).
+    pub fn save(&self) -> Result<()> {
+        let path = Self::data_dir()?.join("config.toml");
+        std::fs::create_dir_all(path.parent().unwrap())?;
+        self.save_to(&path)
+    }
 }
 
 #[cfg(test)]
@@ -138,4 +184,54 @@ batch_size = 128
         let cfg = Config::load().unwrap();
         assert_eq!(cfg.batch_size, 64);
     }
+
+    #[test]
+    fn parse_intelligence_config() {
+        let toml_str = r#"
+intelligence = true
+
+[models]
+embed = "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M-Q8_0.gguf"
+rerank = "hf:ggml-org/Qwen3-Reranker-0.6B-Q8_0-GGUF/qwen3-reranker-0.6b-q8_0.gguf"
+"#;
+        let cfg: Config = toml::from_str(toml_str).unwrap();
+        assert_eq!(cfg.intelligence, Some(true));
+        assert!(cfg.models.embed.is_some());
+        assert!(cfg.models.rerank.is_some());
+        assert!(cfg.models.expand.is_none());
+    }
+
+    #[test]
+    fn intelligence_defaults_to_none() {
+        let cfg = Config::default();
+        assert!(cfg.intelligence.is_none());
+        assert!(cfg.models.embed.is_none());
+    }
+
+    #[test]
+    fn intelligence_false_disables_features() {
+        let toml_str = r#"intelligence = false"#;
+        let cfg: Config = toml::from_str(toml_str).unwrap();
+        assert_eq!(cfg.intelligence, Some(false));
+        assert!(!cfg.intelligence_enabled());
+    }
+
+    #[test]
+    fn test_config_roundtrip_with_intelligence() {
+        let dir = tempfile::tempdir().unwrap();
+        let config_path = dir.path().join("config.toml");
+
+        let mut cfg = Config::default();
+        cfg.intelligence = Some(true);
+        cfg.models.embed = Some("hf:custom/model/embed.gguf".into());
+
+        cfg.save_to(&config_path).unwrap();
+
+        let loaded = Config::load_from(&config_path).unwrap();
+        assert_eq!(loaded.intelligence, Some(true));
+        assert_eq!(
+            loaded.models.embed,
+            Some("hf:custom/model/embed.gguf".into())
+        );
+    }
 }

src/config.rs

@@ -633,7 +633,7 @@ pub fn context_topic_with_search(
     params: &ContextParams,
     topic: &str,
     max_chars: usize,
-    embedder: &mut crate::embedder::Embedder,
+    embedder: &mut impl crate::llm::EmbedModel,
 ) -> Result<ContextBundle> {
     let search_output = crate::search::search_internal(topic, 5, params.store, embedder)?;
     context_topic_from_results(params, topic, &search_output.results, max_chars)