diff --git a/crates/q_cli/src/cli/mod.rs b/crates/q_cli/src/cli/mod.rs
index 49bcf6003..51bbf3831 100644
--- a/crates/q_cli/src/cli/mod.rs
+++ b/crates/q_cli/src/cli/mod.rs
@@ -65,6 +65,7 @@ use fig_util::{
     system_info,
 };
 use internal::InternalSubcommand;
+use macos_utils::bundle::get_bundle_path_for_executable;
 use serde::Serialize;
 use tokio::signal::ctrl_c;
 use tracing::{
@@ -195,15 +196,16 @@ pub enum CliRootCommands {
     /// Open the dashboard
     Dashboard,
     /// AI assistant in your terminal
+    #[command(disable_help_flag = true)]
     Chat {
-        /// Args for the chat command
+        /// Args for the chat subcommand
         #[arg(trailing_var_arg = true, allow_hyphen_values = true)]
         args: Vec<String>,
     },
     /// Model Context Protocol (MCP)
     #[command(disable_help_flag = true)]
     Mcp {
-        /// Args for the MCP subcommand (passed through to `qchat mcp …`)
+        /// Args for the MCP subcommand
         #[arg(trailing_var_arg = true, allow_hyphen_values = true)]
         args: Vec<String>,
     },
@@ -352,11 +354,18 @@ impl Cli {
                 CliRootCommands::Telemetry(subcommand) => subcommand.execute().await,
                 CliRootCommands::Version { changelog } => Self::print_version(changelog),
                 CliRootCommands::Dashboard => launch_dashboard(false).await,
-                CliRootCommands::Chat { args } => Self::execute_chat("chat", Some(args), true).await,
+                CliRootCommands::Chat { args } => {
+                    if args.iter().any(|arg| ["--help", "-h"].contains(&arg.as_str())) {
+                        return Self::execute_chat("chat", Some(vec!["--help".to_owned()]), false).await;
+                    }
+
+                    Self::execute_chat("chat", Some(args), true).await
+                },
                 CliRootCommands::Mcp { args } => {
                     if args.iter().any(|arg| ["--help", "-h"].contains(&arg.as_str())) {
                         return Self::execute_chat("mcp", Some(vec!["--help".to_owned()]), false).await;
                     }
+
                     Self::execute_chat("mcp", Some(args), true).await
                 },
                 CliRootCommands::Inline(subcommand) => subcommand.execute(&cli_context).await,
@@ -367,6 +376,14 @@ impl Cli {
     }
 
     pub async fn execute_chat(subcmd: &str, args: Option<Vec<String>>, enforce_login: bool) -> Result<ExitCode> {
+        cfg_if::cfg_if! {
+            if #[cfg(target_os = "macos")] {
+                let path = get_bundle_path_for_executable(CHAT_BINARY_NAME).unwrap_or(home_local_bin()?.join(CHAT_BINARY_NAME));
+            } else {
+                let path = home_local_bin()?.join(CHAT_BINARY_NAME);
+            }
+        }
+
         if enforce_login {
             assert_logged_in().await?;
         }
@@ -382,7 +399,7 @@ impl Cli {
             }
         }
 
-        let mut cmd = tokio::process::Command::new(home_local_bin()?.join(CHAT_BINARY_NAME));
+        let mut cmd = tokio::process::Command::new(&path);
         cmd.arg(subcmd);
         if let Some(args) = args {
             cmd.args(args);
diff --git a/crates/q_cli/src/main.rs b/crates/q_cli/src/main.rs
index 9f5b3e9ea..54d059acf 100644
--- a/crates/q_cli/src/main.rs
+++ b/crates/q_cli/src/main.rs
@@ -33,19 +33,6 @@ fn main() -> Result<ExitCode> {
         Some("init" | "_" | "internal" | "completion" | "hook" | "chat")
     );
 
-    let runtime = if multithread {
-        tokio::runtime::Builder::new_multi_thread()
-    } else {
-        tokio::runtime::Builder::new_current_thread()
-    }
-    .enable_all()
-    .build()?;
-
-    // Hack as clap doesn't expose a custom command help.
-    if subcommand.as_deref() == Some("chat") && args.any(|arg| ["--help", "-h"].contains(&arg.as_str())) {
-        runtime.block_on(cli::Cli::execute_chat("chat", Some(vec!["--help".to_owned()]), true))?;
-    }
-
     let parsed = match cli::Cli::try_parse() {
         Ok(cli) => cli,
         Err(err) => {
@@ -73,6 +60,14 @@ fn main() -> Result<ExitCode> {
 
     let verbose = parsed.verbose > 0;
 
+    let runtime = if multithread {
+        tokio::runtime::Builder::new_multi_thread()
+    } else {
+        tokio::runtime::Builder::new_current_thread()
+    }
+    .enable_all()
+    .build()?;
+
     let result = runtime.block_on(async {
         let result = parsed.execute().await;
         fig_telemetry::finish_telemetry().await;