feat(llm): support more llms & clean the workspace

Cargo.toml

@@ -1,6 +1,6 @@
 [workspace]
 resolver = "3"
-members = ["app/cli", "app/client", "app/gateway", "app/hub", "app/protocol", "crates/*", "crates/llm/deepseek"]
+members = ["app/*", "crates/*", "llm/*"]
 
 [workspace.package]
 version = "0.0.9"
@@ -11,14 +11,17 @@ repository = "https://github.com/clearloop/walrus"
 keywords = ["llm", "agent", "ai"]
 
 [workspace.dependencies]
-deepseek = { path = "crates/llm/deepseek", package = "walrus-deepseek", version = "0.0.9" }
-llm = { path = "crates/llm", package = "walrus-llm", version = "0.0.9" }
-agent = { path = "crates/core", package = "walrus-core", version = "0.0.9" }
+deepseek = { path = "llm/deepseek", package = "walrus-deepseek", version = "0.0.9" }
+openai = { path = "llm/openai", package = "walrus-openai", version = "0.0.9" }
+claude = { path = "llm/claude", package = "walrus-claude", version = "0.0.9" }
+mistral = { path = "llm/mistral", package = "walrus-mistral", version = "0.0.9" }
+llm = { path = "crates/llm", package = "walrus-llm", version = "0.0.9", default-features = true }
+wcore = { path = "crates/core", package = "walrus-core", version = "0.0.9" }
 runtime = { path = "crates/runtime", package = "walrus-runtime", version = "0.0.9" }
 sqlite = { path = "crates/sqlite", package = "walrus-sqlite", version = "0.0.9" }
 telegram = { path = "crates/telegram", package = "walrus-telegram", version = "0.0.9" }
 protocol = { path = "app/protocol", package = "walrus-protocol", version = "0.0.9" }
-gateway = { path = "app/gateway", package = "walrus-gateway", version = "0.0.9" }
+daemon = { path = "app/daemon", package = "walrus-daemon", version = "0.0.9" }
 client = { path = "app/client", package = "walrus-client", version = "0.0.9" }
 
 # crates.io
@@ -36,7 +39,13 @@ serde_json = "1"
 serde_yaml = "0.9"
 smallvec = { version = "1", features = ["serde"] }
 tempfile = "3"
-tokio = { version = "1", features = ["rt-multi-thread", "macros", "signal", "io-util", "net"] }
+tokio = { version = "1", features = [
+    "rt-multi-thread",
+    "macros",
+    "signal",
+    "io-util",
+    "net",
+] }
 tracing = "0.1"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }
 dotenvy = "0.15"

app/gateway/Cargo.toml → app/daemon/Cargo.toml

@@ -1,5 +1,5 @@
 [package]
-name = "walrus-gateway"
+name = "walrus-daemon"
 version.workspace = true
 edition.workspace = true
 authors.workspace = true
@@ -13,8 +13,11 @@ path = "src/bin/main.rs"
 
 [dependencies]
 runtime = { workspace = true }
-agent = { workspace = true }
+wcore = { workspace = true }
 deepseek = { workspace = true }
+openai = { workspace = true }
+claude = { workspace = true }
+mistral = { workspace = true }
 sqlite = { workspace = true }
 llm = { workspace = true }
 protocol = { workspace = true }

app/gateway/src/bin/main.rs → app/daemon/src/bin/main.rs

@@ -6,7 +6,7 @@
 use anyhow::Result;
 use tokio::signal;
 use tracing_subscriber::EnvFilter;
-use walrus_gateway::config;
+use walrus_daemon::config;
 
 #[tokio::main]
 async fn main() -> Result<()> {
@@ -20,7 +20,7 @@ async fn main() -> Result<()> {
         tracing::info!("created config directory at {}", config_dir.display());
     }
 
-    let handle = walrus_gateway::serve(&config_dir, None).await?;
+    let handle = walrus_daemon::serve(&config_dir, None).await?;
     tracing::info!("walrusd listening on {}", handle.socket_path.display());
 
     signal::ctrl_c().await?;

app/gateway/src/channel/router.rs → app/daemon/src/channel/router.rs

@@ -4,8 +4,8 @@
 //! and channel ID with three-tier fallback: exact match, platform
 //! catch-all, default agent (DD#3).
 
-use agent::Platform;
 use compact_str::CompactString;
+use wcore::Platform;
 
 /// A routing rule mapping platform/channel to an agent.
 #[derive(Debug, Clone)]

app/gateway/src/config.rs → app/daemon/src/config.rs

@@ -55,21 +55,53 @@ pub struct ServerConfig {
 /// LLM provider configuration.
 #[derive(Debug, Serialize, Deserialize)]
 pub struct LlmConfig {
+    /// Which LLM provider to use.
+    #[serde(default)]
+    pub provider: ProviderKind,
     /// Model identifier.
     pub model: CompactString,
     /// API key (supports `${ENV_VAR}` expansion).
+    #[serde(default)]
     pub api_key: String,
+    /// Optional base URL override for the provider endpoint.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub base_url: Option<String>,
 }
 
 impl Default for LlmConfig {
     fn default() -> Self {
         Self {
+            provider: ProviderKind::DeepSeek,
             model: "deepseek-chat".into(),
             api_key: "${DEEPSEEK_API_KEY}".to_owned(),
+            base_url: None,
         }
     }
 }
 
+/// Supported LLM provider kinds.
+#[derive(Debug, Default, Serialize, Deserialize, Clone, Copy, PartialEq, Eq)]
+#[serde(rename_all = "snake_case")]
+pub enum ProviderKind {
+    /// DeepSeek API (default).
+    #[default]
+    DeepSeek,
+    /// OpenAI API.
+    OpenAI,
+    /// Grok (xAI) API — OpenAI-compatible.
+    Grok,
+    /// Qwen (Alibaba DashScope) API — OpenAI-compatible.
+    Qwen,
+    /// Kimi (Moonshot) API — OpenAI-compatible.
+    Kimi,
+    /// Ollama local API — OpenAI-compatible, no key required.
+    Ollama,
+    /// Claude (Anthropic) Messages API.
+    Claude,
+    /// Mistral chat completions API.
+    Mistral,
+}
+
 /// Memory backend configuration.
 #[derive(Debug, Default, Serialize, Deserialize)]
 #[serde(default)]

app/gateway/src/feature/memory.rs → app/daemon/src/feature/memory.rs

@@ -3,10 +3,10 @@
 //! Wraps [`InMemory`] and [`SqliteMemory<NoEmbedder>`] with Memory trait
 //! delegation, following the Provider enum pattern (DD#22).
 
-use agent::{InMemory, Memory, MemoryEntry, NoEmbedder, RecallOptions};
 use anyhow::Result;
 use sqlite::SqliteMemory;
 use std::future::Future;
+use wcore::{InMemory, Memory, MemoryEntry, NoEmbedder, RecallOptions};
 
 /// Memory backend selected from configuration.
 ///