diff --git a/local/ai-commit/lua/ai-commit/generator.lua b/local/ai-commit/lua/ai-commit/generator.lua
--- a/local/ai-commit/lua/ai-commit/generator.lua
+++ b/local/ai-commit/lua/ai-commit/generator.lua
@@ -135,6 +135,11 @@ if job then
         table.insert(http_jobs, job)
       end
     end,
+    on_request_start = function()
+      if not done and not aborted then
+        spinner_ui.start_stream_section(spinner)
+      end
+    end,
     on_chunk = function(chunk)
       if not done and not aborted then
         spinner_ui.append_stream(spinner, chunk)
diff --git a/local/ai-commit/lua/ai-commit/providers.lua b/local/ai-commit/lua/ai-commit/providers.lua
--- a/local/ai-commit/lua/ai-commit/providers.lua
+++ b/local/ai-commit/lua/ai-commit/providers.lua
@@ -125,6 +125,8 @@ if status.phase == 'loading' then
       status_callback(action and (action .. ' with ' .. status_model) or ('Loading model ' .. status_model))
     elseif status.phase == 'loaded' then
       status_callback('Loaded model ' .. status_model)
+    elseif status.phase == 'context' then
+      status_callback('Loading prompt context with ' .. status_model)
     elseif status.phase == 'thinking' then
       local suffix = status.tokens_per_second and string.format(' (%.1f t/s)', status.tokens_per_second) or ''
       status_callback((action or 'Thinking') .. ' with ' .. status_model .. suffix)
@@ -134,6 +136,10 @@ status_callback((action or 'Generating response') .. ' with ' .. status_model .. suffix)
     elseif status.phase == 'error' then
       status_callback('Provider error from ' .. status_model)
     end
+  end
+
+  if request_context and request_context.on_request_start then
+    request_context.on_request_start(source_id, model)
   end
 
   ai_provider.chat(provider, {
diff --git a/local/ai-commit/lua/ai-commit/spinner.lua b/local/ai-commit/lua/ai-commit/spinner.lua
--- a/local/ai-commit/lua/ai-commit/spinner.lua
+++ b/local/ai-commit/lua/ai-commit/spinner.lua
@@ -6,6 +6,72 @@ local M = {}
 
 local frames = { '⠋', '⠙', '⠹', '⠸', '⠼', '⠴', '⠦', '⠧', '⠇', '⠏' }
 
+local function preview_width(bufnr)
+  local win = vim.fn.bufwinid(bufnr)
+  if win ~= -1 then
+    return math.max(20, vim.api.nvim_win_get_width(win) - 4)
+  end
+  return 76
+end
+
+local function split_long_token(token, width)
+  local parts = {}
+  local current = ''
+  local index = 0
+  local length = vim.fn.strchars(token)
+
+  while index < length do
+    local char = vim.fn.strcharpart(token, index, 1)
+    if current ~= '' and vim.fn.strdisplaywidth(current .. char) > width then
+      table.insert(parts, current)
+      current = char
+    else
+      current = current .. char
+    end
+    index = index + 1
+  end
+
+  if current ~= '' then
+    table.insert(parts, current)
+  end
+  return parts
+end
+
+local function wrap_line(line, width)
+  if line == '' or vim.fn.strdisplaywidth(line) <= width then
+    return { line }
+  end
+
+  local wrapped = {}
+  local current = ''
+
+  for token in line:gmatch '%S+%s*' do
+    local candidate = current .. token
+    if current ~= '' and vim.fn.strdisplaywidth(candidate) > width then
+      table.insert(wrapped, (current:gsub('%s+$', '')))
+      current = token
+    else
+      current = candidate
+    end
+
+    if vim.fn.strdisplaywidth(current) > width then
+      local parts = split_long_token(current:gsub('%s+$', ''), width)
+      for index, part in ipairs(parts) do
+        if index < #parts then
+          table.insert(wrapped, part)
+        else
+          current = part
+        end
+      end
+    end
+  end
+
+  if current ~= '' then
+    table.insert(wrapped, (current:gsub('%s+$', '')))
+  end
+  return #wrapped > 0 and wrapped or { line }
+end
+
 local function stop_timer_safe(spinner)
   local timer = spinner and spinner.timer
   if not timer then
@@ -25,13 +91,26 @@ end)
 end
 
 ---@param spinner table
+---@param bufnr integer
 ---@return table
-local function preview_virt_lines(spinner)
+local function preview_virt_lines(spinner, bufnr)
   local lines = {}
+  local width = preview_width(bufnr)
   for _, line in ipairs(spinner.stream_preview) do
-    table.insert(lines, { { line, 'Comment' } })
+    for _, wrapped in ipairs(wrap_line(line, width)) do
+      table.insert(lines, wrapped)
+    end
   end
-  return lines
+
+  while #lines > (config.values.preview_lines or 5) do
+    table.remove(lines, 1)
+  end
+
+  local virt_lines = {}
+  for _, line in ipairs(lines) do
+    table.insert(virt_lines, { { line, 'Comment' } })
+  end
+  return virt_lines
 end
 
 ---@param bufnr integer
@@ -77,7 +156,7 @@ right_gravity = false,
       virt_text = virt_text,
       virt_text_pos = 'eol',
     }
-    local virt_lines = preview_virt_lines(spinner)
+    local virt_lines = preview_virt_lines(spinner, bufnr)
     if #virt_lines > 0 then
       opts.virt_lines = virt_lines
       opts.virt_lines_above = false
@@ -113,8 +192,19 @@ end
     spinner.stream_preview[#spinner.stream_preview] = (spinner.stream_preview[#spinner.stream_preview] or '') .. chunk
   end
 
-  while #spinner.stream_preview > (config.values.preview_lines or 5) do
+  while #spinner.stream_preview > 200 do
     table.remove(spinner.stream_preview, 1)
+  end
+end
+
+---@param spinner table|nil
+function M.start_stream_section(spinner)
+  if not spinner or #spinner.stream_preview == 0 then
+    return
+  end
+
+  if spinner.stream_preview[#spinner.stream_preview] ~= '' then
+    table.insert(spinner.stream_preview, '')
   end
 end
 
diff --git a/local/ai-provider/lua/ai-provider/init.lua b/local/ai-provider/lua/ai-provider/init.lua
--- a/local/ai-provider/lua/ai-provider/init.lua
+++ b/local/ai-provider/lua/ai-provider/init.lua
@@ -44,7 +44,7 @@ ---@field eval_duration? integer Generation duration in nanoseconds, when reported.
 
 ---@class AiProviderStatus
 ---@field provider string Provider name.
----@field phase string Standard phase, for example `loading`, `loaded`, `thinking`, `generating`, `done`, or `error`.
+---@field phase string Standard phase, for example `loading`, `loaded`, `context`, `thinking`, `generating`, `done`, or `error`.
 ---@field message string Human-readable status message.
 ---@field model? string Model/profile name.
 ---@field used_model? string Raw provider model name.
diff --git a/local/ai-provider/lua/ai-provider/providers/ollama.lua b/local/ai-provider/lua/ai-provider/providers/ollama.lua
--- a/local/ai-provider/lua/ai-provider/providers/ollama.lua
+++ b/local/ai-provider/lua/ai-provider/providers/ollama.lua
@@ -295,14 +295,7 @@ return
         end
 
         if code ~= 0 then
-          log.error(
-            'ollama request process failed code='
-              .. tostring(code)
-              .. ' model='
-              .. tostring(raw_model)
-              .. ' error='
-              .. tostring(error_message)
-          )
+          log.error('ollama request process failed code=' .. tostring(code) .. ' model=' .. tostring(raw_model) .. ' error=' .. tostring(error_message))
           if request.callback then
             request.callback(nil, {
               requested_model = selected_model,
@@ -354,12 +347,12 @@ log.error('ollama provider error requested_model=' .. tostring(selected_model) .. ' error=' .. provider_error)
           if request.callback then
             request.callback(nil, meta)
           end
-        emit_status {
-          phase = 'error',
-          message = provider_error,
-          tokens = metrics.eval_count,
-          tokens_per_second = tokens_per_second(metrics.eval_count, metrics.eval_duration),
-        }
+          emit_status {
+            phase = 'error',
+            message = provider_error,
+            tokens = metrics.eval_count,
+            tokens_per_second = tokens_per_second(metrics.eval_count, metrics.eval_duration),
+          }
           return
         end
         if done_reason == 'length' then
@@ -467,14 +460,7 @@ end
 
         if response.status ~= 200 then
           local error_message = response.error or response.body or 'ollama preload failed'
-          log.error(
-            'ollama preload failed status='
-              .. tostring(response.status)
-              .. ' model='
-              .. tostring(raw_model)
-              .. ' error='
-              .. tostring(error_message)
-          )
+          log.error('ollama preload failed status=' .. tostring(response.status) .. ' model=' .. tostring(raw_model) .. ' error=' .. tostring(error_message))
           if request.callback then
             request.callback(nil, {
               requested_model = selected_model,
@@ -506,8 +492,8 @@ load_duration and string.format('%.0f', load_duration / 1e6) or 'nil'
           )
         )
         emit_status {
-          phase = 'generating',
-          message = 'Generating response',
+          phase = 'context',
+          message = 'Loading prompt context',
         }
         job = run_chat()
         if request.register_http_job then