refactoring folders

lib/elixir_ai_web/features/admin/admin_live.ex

@@ -1,108 +1,129 @@
 defmodule ElixirAiWeb.AdminLive do
+  import ElixirAi.PubsubTopics
   use ElixirAiWeb, :live_view
   require Logger
 
-  @refresh_ms 1_000
-
   def mount(_params, _session, socket) do
-    if connected?(socket) do
-      :net_kernel.monitor_nodes(true)
-      :pg.join(ElixirAi.LiveViewPG, {:liveview, __MODULE__}, self())
-      schedule_refresh()
-    end
+    socket =
+      if connected?(socket) do
+        :net_kernel.monitor_nodes(true)
+        # Join before monitoring so our own join doesn't trigger a spurious refresh.
+        :pg.join(ElixirAi.LiveViewPG, {:liveview, __MODULE__}, self())
+        {pg_ref, _} = :pg.monitor_scope(ElixirAi.LiveViewPG)
+        {runner_pg_ref, _} = :pg.monitor_scope(ElixirAi.RunnerPG)
+        {singleton_pg_ref, _} = :pg.monitor_scope(ElixirAi.SingletonPG)
 
-    {:ok, assign(socket, cluster_info: gather_info())}
+        socket
+        |> assign(pg_ref: pg_ref)
+        |> assign(runner_pg_ref: runner_pg_ref)
+        |> assign(singleton_pg_ref: singleton_pg_ref)
+      else
+        assign(socket, pg_ref: nil, runner_pg_ref: nil, singleton_pg_ref: nil)
+      end
+
+    {:ok,
+     socket
+     |> assign(nodes: gather_node_statuses())
+     |> assign(singleton_locations: gather_singleton_locations())
+     |> assign(chat_runners: gather_chat_runners())
+     |> assign(liveviews: gather_liveviews())}
   end
 
   def handle_info({:nodeup, _node}, socket) do
-    {:noreply, assign(socket, cluster_info: gather_info())}
+    {:noreply, assign(socket, nodes: gather_node_statuses())}
   end
 
   def handle_info({:nodedown, _node}, socket) do
-    {:noreply, assign(socket, cluster_info: gather_info())}
+    {:noreply, assign(socket, nodes: gather_node_statuses())}
   end
 
-  def handle_info(:refresh, socket) do
-    schedule_refresh()
-    {:noreply, assign(socket, cluster_info: gather_info())}
+  def handle_info(:refresh_singletons, socket) do
+    {:noreply, assign(socket, singleton_locations: gather_singleton_locations())}
   end
 
-  defp schedule_refresh, do: Process.send_after(self(), :refresh, @refresh_ms)
+  def handle_info({ref, change, _group, _pids}, %{assigns: %{singleton_pg_ref: ref}} = socket)
+      when is_reference(ref) and change in [:join, :leave] do
+    {:noreply, assign(socket, singleton_locations: gather_singleton_locations())}
+  end
 
-  defp gather_info do
-    import ElixirAi.PubsubTopics
+  def handle_info({ref, change, _group, _pids}, %{assigns: %{pg_ref: ref}} = socket)
+      when is_reference(ref) and change in [:join, :leave] do
+    {:noreply, assign(socket, liveviews: gather_liveviews())}
+  end
 
+  def handle_info({ref, change, _group, _pids}, %{assigns: %{runner_pg_ref: ref}} = socket)
+      when is_reference(ref) and change in [:join, :leave] do
+    {:noreply, assign(socket, chat_runners: gather_chat_runners())}
+  end
+
+  defp gather_node_statuses do
     all_nodes = [Node.self() | Node.list()]
-    configured = ElixirAi.ClusterSingleton.configured_singletons()
 
-    node_statuses =
-      Enum.map(all_nodes, fn node ->
-        status =
-          if node == Node.self() do
-            try do
-              ElixirAi.ClusterSingleton.status()
-            catch
-              _, _ -> :unreachable
-            end
-          else
-            case :rpc.call(node, ElixirAi.ClusterSingleton, :status, [], 3_000) do
-              {:badrpc, _} -> :unreachable
-              result -> result
-            end
+    Enum.map(all_nodes, fn node ->
+      status =
+        if node == Node.self() do
+          try do
+            ElixirAi.ClusterSingleton.status()
+          catch
+            _, _ -> :unreachable
           end
-
-        {node, status}
-      end)
-
-    singleton_locations =
-      Enum.map(configured, fn module ->
-        location =
-          case Horde.Registry.lookup(ElixirAi.ChatRegistry, module) do
-            [{pid, _}] -> node(pid)
-            _ -> nil
+        else
+          case :rpc.call(node, ElixirAi.ClusterSingleton, :status, [], 3_000) do
+            {:badrpc, _} -> :unreachable
+            result -> result
           end
+        end
 
-        {module, location}
-      end)
+      {node, status}
+    end)
+  end
 
-    # All ChatRunner entries in the distributed registry, keyed by conversation name.
-    # Each entry is a {name, node, pid, supervisor_node} tuple.
-    chat_runners =
-      Horde.DynamicSupervisor.which_children(ElixirAi.ChatRunnerSupervisor)
+  defp gather_singleton_locations do
+    running =
+      :pg.which_groups(ElixirAi.SingletonPG)
       |> Enum.flat_map(fn
-        {_, pid, _, _} when is_pid(pid) ->
-          case Horde.Registry.select(ElixirAi.ChatRegistry, [
-                 {{:"$1", pid, :"$2"}, [], [{{:"$1", pid, :"$2"}}]}
-               ]) do
-            [{name, ^pid, _}] when is_binary(name) -> [{name, node(pid), pid}]
+        {:singleton, module} ->
+          case :pg.get_members(ElixirAi.SingletonPG, {:singleton, module}) do
+            [pid | _] -> [{module, node(pid)}]
             _ -> []
           end
 
         _ ->
           []
       end)
-      |> Enum.sort_by(&elem(&1, 0))
+      |> Map.new()
 
-    # :pg is cluster-wide — one local call returns members from all nodes.
-    # Processes are automatically removed from their group when they die.
-    liveviews =
-      :pg.which_groups(ElixirAi.LiveViewPG)
-      |> Enum.flat_map(fn
-        {:liveview, view} ->
-          :pg.get_members(ElixirAi.LiveViewPG, {:liveview, view})
-          |> Enum.map(fn pid -> {view, node(pid)} end)
+    ElixirAi.ClusterSingleton.configured_singletons()
+    |> Enum.map(fn module -> {module, Map.get(running, module)} end)
+  end
 
-        _ ->
-          []
-      end)
+  # All ChatRunner entries via :pg membership, keyed by conversation name.
+  # Each entry is a {name, node, pid} tuple.
+  defp gather_chat_runners do
+    :pg.which_groups(ElixirAi.RunnerPG)
+    |> Enum.flat_map(fn
+      {:runner, name} ->
+        :pg.get_members(ElixirAi.RunnerPG, {:runner, name})
+        |> Enum.map(fn pid -> {name, node(pid), pid} end)
 
-    %{
-      nodes: node_statuses,
-      configured_singletons: configured,
-      singleton_locations: singleton_locations,
-      chat_runners: chat_runners,
-      liveviews: liveviews
-    }
+      _ ->
+        []
+    end)
+    |> Enum.sort_by(&elem(&1, 0))
+  end
+
+  # :pg is cluster-wide — one local call returns members from all nodes.
+  # Processes are automatically removed from their group when they die.
+  defp gather_liveviews do
+    :pg.which_groups(ElixirAi.LiveViewPG)
+    |> Enum.flat_map(fn
+      {:liveview, view} ->
+        :pg.get_members(ElixirAi.LiveViewPG, {:liveview, view})
+        |> Enum.map(fn pid -> {view, node(pid)} end)
+
+      _ ->
+        []
+    end)
   end
 
   def render(assigns) do
@@ -111,13 +132,13 @@ defmodule ElixirAiWeb.AdminLive do
       <h1 class="text-lg font-semibold text-seafoam-200 tracking-wide">Cluster Admin</h1>
 
       <div class="grid gap-4 grid-cols-1 lg:grid-cols-2 xl:grid-cols-3">
-        <%= for {node, status} <- @cluster_info.nodes do %>
+        <%= for {node, status} <- @nodes do %>
           <% node_singletons =
-            Enum.filter(@cluster_info.singleton_locations, fn {_, loc} -> loc == node end) %>
+            Enum.filter(@singleton_locations, fn {_, loc} -> loc == node end) %>
           <% node_runners =
-            Enum.filter(@cluster_info.chat_runners, fn {_, rnode, _} -> rnode == node end) %>
+            Enum.filter(@chat_runners, fn {_, rnode, _} -> rnode == node end) %>
           <% node_liveviews =
-            @cluster_info.liveviews
+            @liveviews
             |> Enum.filter(fn {_, n} -> n == node end)
             |> Enum.group_by(fn {view, _} -> view end) %>
 
@@ -126,7 +147,9 @@ defmodule ElixirAiWeb.AdminLive do
               <div class="flex items-center gap-2">
                 <span class="font-mono text-sm font-semibold text-seafoam-200">{node}</span>
                 <%= if node == Node.self() do %>
-                  <span class="text-xs bg-seafoam-800/50 text-seafoam-400 px-1.5 py-0.5 rounded">self</span>
+                  <span class="text-xs bg-seafoam-800/50 text-seafoam-400 px-1.5 py-0.5 rounded">
+                    self
+                  </span>
                 <% end %>
               </div>
               <.status_badge status={status} />
@@ -191,7 +214,7 @@ defmodule ElixirAiWeb.AdminLive do
       </div>
 
       <% unlocated =
-        Enum.filter(@cluster_info.singleton_locations, fn {_, loc} -> is_nil(loc) end) %>
+        Enum.filter(@singleton_locations, fn {_, loc} -> is_nil(loc) end) %>
       <%= if unlocated != [] do %>
         <section>
           <h2 class="text-xs font-semibold uppercase tracking-widest text-red-500 mb-2">
@@ -207,7 +230,9 @@ defmodule ElixirAiWeb.AdminLive do
         </section>
       <% end %>
 
-      <p class="text-xs text-seafoam-800">Refreshes every 1s or on node events.</p>
+      <p class="text-xs text-seafoam-800">
+        Nodes, singletons, liveviews &amp; runners all refresh on membership changes.
+      </p>
     </div>
     """
   end

lib/elixir_ai_web/features/chat/stream_handler.ex

@@ -0,0 +1,183 @@
+defmodule ElixirAi.ChatRunner.StreamHandler do
+  require Logger
+  import ElixirAi.ChatRunner.OutboundHelpers
+
+  def handle({:start_new_ai_response, id}, state) do
+    starting_response = %{id: id, reasoning_content: "", content: "", tool_calls: []}
+    broadcast_ui(state.name, {:start_ai_response_stream, starting_response})
+    {:noreply, %{state | streaming_response: starting_response}}
+  end
+
+  def handle({:ai_reasoning_chunk, _id, reasoning_content}, state) do
+    broadcast_ui(state.name, {:reasoning_chunk_content, reasoning_content})
+
+    {:noreply,
+     %{
+       state
+       | streaming_response: %{
+           state.streaming_response
+           | reasoning_content: state.streaming_response.reasoning_content <> reasoning_content
+         }
+     }}
+  end
+
+  def handle({:ai_text_chunk, _id, text_content}, state) do
+    broadcast_ui(state.name, {:text_chunk_content, text_content})
+
+    {:noreply,
+     %{
+       state
+       | streaming_response: %{
+           state.streaming_response
+           | content: state.streaming_response.content <> text_content
+         }
+     }}
+  end
+
+  def handle({:ai_text_stream_finish, _id}, state) do
+    Logger.info(
+      "AI stream finished for id #{state.streaming_response.id}, broadcasting end of AI response"
+    )
+
+    final_message = %{
+      role: :assistant,
+      content: state.streaming_response.content,
+      reasoning_content: state.streaming_response.reasoning_content,
+      tool_calls: state.streaming_response.tool_calls
+    }
+
+    broadcast_ui(state.name, {:end_ai_response, final_message})
+    store_message(state.name, final_message)
+
+    {:noreply,
+     %{
+       state
+       | streaming_response: nil,
+         messages: state.messages ++ [final_message]
+     }}
+  end
+
+  def handle(
+        {:ai_tool_call_start, _id, {tool_name, tool_args_start, tool_index, tool_call_id}},
+        state
+      ) do
+    Logger.info("AI started tool call #{tool_name}")
+
+    new_streaming_response = %{
+      state.streaming_response
+      | tool_calls:
+          state.streaming_response.tool_calls ++
+            [
+              %{
+                name: tool_name,
+                arguments: tool_args_start,
+                index: tool_index,
+                id: tool_call_id
+              }
+            ]
+    }
+
+    {:noreply, %{state | streaming_response: new_streaming_response}}
+  end
+
+  def handle({:ai_tool_call_middle, _id, {tool_args_diff, tool_index}}, state) do
+    new_streaming_response = %{
+      state.streaming_response
+      | tool_calls:
+          Enum.map(state.streaming_response.tool_calls, fn
+            %{arguments: existing_args, index: ^tool_index} = tool_call ->
+              %{tool_call | arguments: existing_args <> tool_args_diff}
+
+            other ->
+              other
+          end)
+    }
+
+    {:noreply, %{state | streaming_response: new_streaming_response}}
+  end
+
+  def handle({:ai_tool_call_end, id}, state) do
+    tool_request_message = %{
+      role: :assistant,
+      content: state.streaming_response.content,
+      reasoning_content: state.streaming_response.reasoning_content,
+      tool_calls: state.streaming_response.tool_calls
+    }
+
+    broadcast_ui(state.name, {:tool_request_message, tool_request_message})
+
+    {failed_call_messages, pending_call_ids} =
+      Enum.reduce(state.streaming_response.tool_calls, {[], []}, fn tool_call,
+                                                                    {failed, pending} ->
+        with {:ok, decoded_args} <- Jason.decode(tool_call.arguments),
+             tool when not is_nil(tool) <-
+               Enum.find(state.server_tools ++ state.liveview_tools ++ state.page_tools, fn t ->
+                 t.name == tool_call.name
+               end) do
+          tool.run_function.(id, tool_call.id, decoded_args)
+          {failed, [tool_call.id | pending]}
+        else
+          {:error, e} ->
+            error_msg = "Failed to decode tool arguments: #{inspect(e)}"
+            Logger.error("Tool call #{tool_call.name} failed: #{error_msg}")
+            {[%{role: :tool, content: error_msg, tool_call_id: tool_call.id} | failed], pending}
+
+          nil ->
+            error_msg = "No tool definition found for #{tool_call.name}"
+            Logger.error(error_msg)
+            {[%{role: :tool, content: error_msg, tool_call_id: tool_call.id} | failed], pending}
+        end
+      end)
+
+    store_message(state.name, [tool_request_message] ++ failed_call_messages)
+
+    {:noreply,
+     %{
+       state
+       | messages: state.messages ++ [tool_request_message] ++ failed_call_messages,
+         pending_tool_calls: pending_call_ids
+     }}
+  end
+
+  def handle({:tool_response, _id, tool_call_id, result}, state) do
+    new_message = %{role: :tool, content: inspect(result), tool_call_id: tool_call_id}
+
+    broadcast_ui(state.name, {:one_tool_finished, new_message})
+    store_message(state.name, new_message)
+
+    new_pending_tool_calls =
+      Enum.filter(state.pending_tool_calls, fn id -> id != tool_call_id end)
+
+    new_streaming_response =
+      case new_pending_tool_calls do
+        [] -> nil
+        _ -> state.streaming_response
+      end
+
+    if new_pending_tool_calls == [] do
+      broadcast_ui(state.name, :tool_calls_finished)
+
+      ElixirAi.ChatUtils.request_ai_response(
+        self(),
+        messages_with_system_prompt(state.messages ++ [new_message], state.system_prompt),
+        state.server_tools ++ state.liveview_tools ++ state.page_tools,
+        state.provider,
+        state.tool_choice
+      )
+    end
+
+    {:noreply,
+     %{
+       state
+       | pending_tool_calls: new_pending_tool_calls,
+         streaming_response: new_streaming_response,
+         messages: state.messages ++ [new_message]
+     }}
+  end
+
+  def handle({:ai_request_error, reason}, state) do
+    Logger.error("AI request error: #{inspect(reason)}")
+    broadcast_ui(state.name, {:ai_request_error, reason})
+    {:noreply, %{state | streaming_response: nil, pending_tool_calls: []}}
+  end
+end

lib/elixir_ai_web/utils/live_view_pg.ex

@@ -0,0 +1,16 @@
+defmodule ElixirAi.LiveViewPG do
+  @moduledoc """
+  Named :pg scope for tracking LiveView processes across the cluster.
+  Each LiveView joins {:liveview, ViewModule} on connect; :pg syncs membership
+  automatically and removes dead processes without any additional cleanup.
+  """
+
+  def child_spec(_opts) do
+    %{
+      id: __MODULE__,
+      start: {:pg, :start_link, [__MODULE__]},
+      type: :worker,
+      restart: :permanent
+    }
+  end
+end

lib/elixir_ai_web/utils/page_tools_pg.ex

@@ -0,0 +1,15 @@
+defmodule ElixirAi.PageToolsPG do
+  @moduledoc """
+  Named :pg scope for tracking LiveViews that implement AiControllable.
+  Group key is `{:page, voice_session_id}` — one group per browser session.
+  """
+
+  def child_spec(_opts) do
+    %{
+      id: __MODULE__,
+      start: {:pg, :start_link, [__MODULE__]},
+      type: :worker,
+      restart: :permanent
+    }
+  end
+end