engram-memory 0.6.0

Core binary and shared infrastructure for Engram memory system
Documentation
use std::path::{Path, PathBuf};
use std::sync::{Arc, Mutex, RwLock};

use tokio::io::{AsyncBufReadExt, AsyncWriteExt, BufReader};
use tokio::net::UnixListener;
use tokio::signal::unix::{SignalKind, signal};

use engram_embeddings::Embedder;
use engram_llm_client::{EmbeddingProvider, TextGenerator};
use engram_router::Router;
use engram_storage::Database;

use crate::config::{self, Config};
use crate::dispatch;
use crate::error::CoreError;
use crate::indexes::IndexSet;
use crate::lock_helpers;
use crate::protocol::{JsonRequest, JsonResponse};

const PROJECT_DB_RELATIVE: &str = ".engram/engram.db";
const PROJECT_SOCKET_RELATIVE: &str = ".engram/engram.sock";

const ROUTER_DEFAULT_ALPHA: f32 = 0.1;
const ROUTER_DEFAULT_EPSILON: f32 = 0.15;

pub struct ServerState {
    pub database: Mutex<Database>,
    pub indexes: RwLock<IndexSet>,
    pub embedder: Embedder,
    pub router: Mutex<Router>,
    pub config: Config,
    pub database_path: String,
    pub embedding_provider: Arc<dyn EmbeddingProvider + Send + Sync>,
    pub text_generator: Option<Arc<dyn TextGenerator + Send + Sync>>,
}

pub async fn run(config: Config) -> Result<(), CoreError> {
    let cwd = std::env::current_dir()
        .map_err(|error| CoreError::SocketError(format!("cwd unavailable: {error}")))?;
    let home = home_dir_or_error()?;
    let project_dir = config::resolve_project_dir(&cwd, None)?;
    // See execute_command in main.rs for why this construction has to run on
    // the blocking pool. reqwest::blocking::ClientBuilder::build internally
    // spins up and drops a current-thread tokio runtime; doing that on the
    // outer multi-threaded runtime's worker panics.
    let state = {
        let config = config.clone();
        let project_dir = project_dir.clone();
        let home = home.clone();
        tokio::task::spawn_blocking(move || initialize_state(&config, &project_dir, &home))
            .await
            .map_err(|error| CoreError::SocketError(error.to_string()))??
    };
    {
        let database = lock_helpers::lock_db(&state);
        crate::migrations::embedding_model_v1::check(
            &database,
            state.embedding_provider.model_name(),
        )?;
    }
    warn_missing_api_keys(&config);
    let shared_state = Arc::new(state);
    let socket_path = resolve_socket_path(&project_dir, &config)?;
    cleanup_stale_socket(&socket_path);
    let listener = bind_listener(&socket_path)?;
    accept_loop(listener, shared_state).await
}

pub fn check_legacy_database(project_dir: &Path, home_dir: &Path) -> Result<(), CoreError> {
    let project_db = project_dir.join(PROJECT_DB_RELATIVE);
    let home_db = home_dir.join(PROJECT_DB_RELATIVE);
    if !project_db.exists() && home_db.exists() {
        return Err(CoreError::LegacyDatabaseDetected {
            legacy_path: home_db.to_string_lossy().to_string(),
            project_path: project_dir.to_string_lossy().to_string(),
        });
    }
    Ok(())
}

pub(crate) fn initialize_state(
    config: &Config,
    project_dir: &Path,
    home_dir: &Path,
) -> Result<ServerState, CoreError> {
    crate::config::validate_dedup_threshold(config.deduplication.threshold)?;
    crate::config::validate_search_config(&config.search)?;
    crate::config::validate_fts_similarity_floor(config.consolidation.fts_similarity_floor)?;
    check_legacy_database(project_dir, home_dir)?;
    let database_path = resolve_database_path(project_dir);
    let database = Database::open(&database_path)?;
    crate::migrations::run_pending(&database)?;
    let embedding_provider: Arc<dyn EmbeddingProvider + Send + Sync> =
        Arc::from(config.build_embedding_provider()?);
    let text_generator: Option<Arc<dyn TextGenerator + Send + Sync>> =
        match config.build_text_generator() {
            Ok(generator) => Some(Arc::from(generator)),
            Err(error) => {
                eprintln!("warning: text generator unavailable, running in degraded mode: {error}");
                None
            }
        };
    let hnsw_config = config.clone();
    let indexes = crate::persistence::load_or_rebuild(
        &resolve_index_directory(&database_path),
        &database,
        || hnsw_config.build_hnsw_params(),
    )?;
    let embedder = Embedder::new(config.embedding.hyde_threshold);
    let router = Router::new(ROUTER_DEFAULT_ALPHA, ROUTER_DEFAULT_EPSILON);
    Ok(ServerState {
        database: Mutex::new(database),
        indexes: RwLock::new(indexes),
        embedder,
        router: Mutex::new(router),
        config: config.clone(),
        database_path,
        embedding_provider,
        text_generator,
    })
}

fn resolve_database_path(project_dir: &Path) -> String {
    if let Ok(value) = std::env::var("ENGRAM_DB_PATH") {
        return value;
    }
    project_dir
        .join(PROJECT_DB_RELATIVE)
        .to_string_lossy()
        .to_string()
}

fn resolve_socket_path(project_dir: &Path, config: &Config) -> Result<String, CoreError> {
    if let Ok(value) = std::env::var("ENGRAM_SOCKET_PATH") {
        return Ok(value);
    }
    // Prefer project socket when the `.engram/` dir exists (walk-up already validated this).
    if project_dir.join(".engram").is_dir() {
        return Ok(project_dir
            .join(PROJECT_SOCKET_RELATIVE)
            .to_string_lossy()
            .to_string());
    }
    match config.server.socket_path.as_deref() {
        Some(fallback) => Ok(expand_tilde(fallback)),
        None => Err(CoreError::SocketError(
            "no socket path available: no project .engram/ marker found, \
             ENGRAM_SOCKET_PATH not set, and server.socket_path not configured"
                .into(),
        )),
    }
}

fn home_dir_or_error() -> Result<PathBuf, CoreError> {
    match std::env::var("HOME") {
        Ok(value) if !value.is_empty() => Ok(PathBuf::from(value)),
        _ => Err(CoreError::InitFailed(
            "HOME environment variable not set".into(),
        )),
    }
}

pub(crate) fn resolve_index_directory(database_path: &str) -> String {
    std::path::Path::new(database_path)
        .parent()
        .map(|parent| parent.to_string_lossy().to_string())
        .unwrap_or_else(|| ".".to_string())
}

fn cleanup_stale_socket(socket_path: &str) {
    let _ = std::fs::remove_file(socket_path);
}

fn bind_listener(socket_path: &str) -> Result<UnixListener, CoreError> {
    if let Some(parent) = std::path::Path::new(socket_path).parent() {
        std::fs::create_dir_all(parent)
            .map_err(|error| CoreError::SocketError(error.to_string()))?;
    }
    UnixListener::bind(socket_path).map_err(|error| CoreError::SocketError(error.to_string()))
}

async fn accept_loop(listener: UnixListener, state: Arc<ServerState>) -> Result<(), CoreError> {
    let mut sigterm =
        signal(SignalKind::terminate()).map_err(|e| CoreError::SocketError(e.to_string()))?;
    spawn_background_reindex(Arc::clone(&state));
    loop {
        tokio::select! {
            accept_result = listener.accept() => {
                let (stream, _address) = accept_result
                    .map_err(|error| CoreError::SocketError(error.to_string()))?;
                let client_state = Arc::clone(&state);
                tokio::spawn(async move {
                    if let Err(error) = handle_client(stream, client_state).await {
                        eprintln!("client error: {error}");
                    }
                });
            }
            _ = sigterm.recv() => {
                return save_indexes_on_shutdown(&state);
            }
        }
    }
}

fn spawn_background_reindex(state: Arc<ServerState>) {
    let interval_secs = state.config.server.reindex_interval_secs;
    if interval_secs == 0 {
        return;
    }
    tokio::spawn(async move {
        let mut interval = tokio::time::interval(std::time::Duration::from_secs(interval_secs));
        interval.tick().await; // skip first immediate tick
        loop {
            interval.tick().await;
            let reindex_state = Arc::clone(&state);
            let _ = tokio::task::spawn_blocking(move || {
                reindex_unindexed_memories(&reindex_state);
            })
            .await;
        }
    });
}

/// Reindex unindexed memories from SQLite into the HNSW index.
///
/// **Unstable:** exposed for latency benchmarking only
/// (`crates/engram-core/tests/search_latency_bench.rs`). Not part of the public
/// API contract; may change or be removed without notice.
#[doc(hidden)]
pub fn reindex_unindexed_memories(state: &ServerState) {
    let database = lock_helpers::lock_db(state);
    let unindexed = match query_unindexed_ids(&database) {
        Ok(ids) => ids,
        Err(_) => return,
    };
    if unindexed.is_empty() {
        return;
    }
    let mut indexes = lock_helpers::write_indexes(state);
    for memory_id in &unindexed {
        let _ = reindex_single_memory(&database, &mut indexes, memory_id);
    }
}

fn query_unindexed_ids(database: &Database) -> Result<Vec<String>, CoreError> {
    let mut statement = database
        .connection()
        .prepare(
            "SELECT id FROM memories WHERE indexed = 0 \
             AND embedding_context IS NOT NULL LIMIT 100",
        )
        .map_err(|e| CoreError::Storage(engram_storage::StorageError::from(e)))?;
    let rows = statement
        .query_map([], |row| row.get::<_, String>(0))
        .map_err(|e| CoreError::Storage(engram_storage::StorageError::from(e)))?;
    let ids: Vec<String> = rows.flatten().collect();
    Ok(ids)
}

fn reindex_single_memory(
    database: &Database,
    indexes: &mut IndexSet,
    memory_id: &str,
) -> Result<(), CoreError> {
    let memory = database.get_memory(memory_id)?;
    let embedding = crate::persistence::extract_embeddings_from_memory(&memory)?;
    let id_hash = crate::persistence::hash_string_to_u64(memory_id);
    let rng_value = crate::persistence::deterministic_rng(id_hash);
    if !indexes.contains(id_hash) {
        indexes.insert_atomic(id_hash, memory_id, &embedding, rng_value)?;
    }
    database.set_memory_indexed(memory_id, true)?;
    Ok(())
}

async fn handle_client(
    stream: tokio::net::UnixStream,
    state: Arc<ServerState>,
) -> Result<(), CoreError> {
    let (reader, mut writer) = stream.into_split();
    let mut lines = BufReader::new(reader).lines();
    while let Some(line) = lines
        .next_line()
        .await
        .map_err(|error| CoreError::SocketError(error.to_string()))?
    {
        let response = process_request_line(&line, &state).await;
        let serialized = serde_json::to_string(&response)
            .map_err(|error| CoreError::SocketError(error.to_string()))?;
        writer
            .write_all(serialized.as_bytes())
            .await
            .map_err(|error| CoreError::SocketError(error.to_string()))?;
        writer
            .write_all(b"\n")
            .await
            .map_err(|error| CoreError::SocketError(error.to_string()))?;
    }
    Ok(())
}

async fn process_request_line(line: &str, state: &Arc<ServerState>) -> JsonResponse {
    let request: JsonRequest = match serde_json::from_str(line) {
        Ok(parsed) => parsed,
        Err(error) => {
            return JsonResponse::error(String::new(), 4000, format!("invalid json: {error}"));
        }
    };
    let request_id = request.id.clone();
    match dispatch::route(&request.method, state, request.params).await {
        Ok(data) => JsonResponse::success(request_id, data),
        Err(error) => JsonResponse::error(request_id, error_code(&error), error.to_string()),
    }
}

fn error_code(error: &CoreError) -> u32 {
    match error {
        CoreError::ConfigNotFound => 6001,
        CoreError::ConfigParseError(_) => 6002,
        CoreError::InvalidProvider(_) => 6003,
        CoreError::IndexCorrupted(_) => 6004,
        CoreError::RebuildFailed(_) => 6005,
        CoreError::SocketError(_) => 6006,
        CoreError::DispatchError(_) => 6007,
        CoreError::ConfigReadOnly => 6008,
        CoreError::ExportFailed(_) => 6009,
        CoreError::ImportVersionMismatch(_) => 6010,
        CoreError::ImportFailed(_) => 6011,
        CoreError::InitFailed(_) => 6012,
        CoreError::TrainerFailed(_) => 6013,
        CoreError::TrainerTimeout => 6014,
        CoreError::TrainerMalformedOutput(_) => 6015,
        CoreError::ProjectDirNotFound => 6016,
        CoreError::LegacyDatabaseDetected { .. } => 6017,
        CoreError::MigrationSourceNotFound => 6018,
        CoreError::MigrationFailed(_) => 6019,
        CoreError::EmbeddingModelMismatch { .. } => 6020,
        CoreError::IndexHashCollision { .. } => 6021,
        CoreError::ConfigValidation(_) => 6022,
        CoreError::Storage(_) => 1000,
        CoreError::Hnsw(_) => 3000,
        CoreError::Api(_) => 2000,
        CoreError::Consolidation(_) => 5000,
    }
}

fn save_indexes_on_shutdown(state: &Arc<ServerState>) -> Result<(), CoreError> {
    let index_directory = resolve_index_directory(&state.database_path);
    let indexes = lock_helpers::read_indexes(state);
    crate::persistence::save_to_disk(&index_directory, &indexes)
}

fn warn_missing_api_keys(config: &Config) {
    let has_embedding_key = config
        .embedding
        .api_key
        .as_deref()
        .is_some_and(|k| !k.is_empty());
    let has_llm_key = config.llm.api_key.as_deref().is_some_and(|k| !k.is_empty());

    if !has_embedding_key
        && config.embedding.provider != "deterministic"
        && config.embedding.provider != "ollama"
    {
        eprintln!(
            "warning: {} embedding provider configured without api key (set ENGRAM_VOYAGE_API_KEY)",
            config.embedding.provider
        );
    }
    if !has_llm_key && config.llm.provider != "local" && config.llm.provider != "ollama" {
        eprintln!(
            "warning: {} llm provider configured without api key (set ENGRAM_OPENAI_API_KEY)",
            config.llm.provider
        );
    }
}

fn expand_tilde(path: &str) -> String {
    if let Some(rest) = path.strip_prefix("~/")
        && let Ok(home) = std::env::var("HOME")
    {
        return format!("{home}/{rest}");
    }
    path.to_string()
}