diff --git a/renderer.lua b/renderer.lua
index 1e1ff67..20d4a43 100644
--- a/renderer.lua
+++ b/renderer.lua
@@ -1,7 +1,10 @@
 -- renderer.lua — output formatting and ANSI sequences.
 -- Phase 0: assistant text plain-printed with `CMD: ` lines highlighted;
--- exec output framed with the exit code on the closing rule. Syntax
--- highlighting hooks land in Phase 6 (was Phase 5 pre-MCP renumber).
+-- exec output framed with the exit code on the closing rule.
+-- Phase 1: assistant_delta + assistant_flush for streaming render. CMD:
+-- highlighting in streaming mode is deferred (Q12); deltas print raw, the
+-- §6 substrate `CMD: ` line is still extractable by executor afterwards.
+-- Syntax highlighting hooks land in Phase 6 (was Phase 5 pre-MCP renumber).
 
 local M = {}
 
@@ -48,4 +51,26 @@ function M.status(line)
     emit(A.dim, "[aish] ", tostring(line), A.reset, "\n")
 end
 
+-- Streaming assistant output. Phase 1: deltas are written raw — the §6 CMD:
+-- highlighting from M.assistant() is not applied incrementally because
+-- mid-line cursor manipulation isn't worth the complexity for Phase 1.
+-- Q12 (PHASE1.md §10) tracks the upgrade. The full assistant text is still
+-- captured by repl.lua and CMD: extraction works against the reassembled
+-- string after the stream ends.
+
+local stream_buf = nil   -- non-nil while a stream is in progress
+
+function M.assistant_delta(chunk)
+    if not chunk or chunk == "" then return end
+    if stream_buf == nil then stream_buf = "" end
+    stream_buf = stream_buf .. chunk
+    emit(chunk)
+end
+
+function M.assistant_flush()
+    if stream_buf == nil then return end
+    if not stream_buf:match("\n$") then emit("\n") end
+    stream_buf = nil
+end
+
 return M
diff --git a/repl.lua b/repl.lua
index d76b2f8..e398022 100644
--- a/repl.lua
+++ b/repl.lua
@@ -68,21 +68,30 @@ function M.run(config)
         end
     end
 
-    -- Send user text to the active model, render the response, and (per
-    -- §6 + config.shell.confirm_cmd) optionally execute extracted CMD: lines.
+    -- Send user text to the active model, render the response token-by-token
+    -- via broker.chat_stream, and (per §6 + config.shell.confirm_cmd) optionally
+    -- execute extracted CMD: lines on the reassembled full text.
     local function ask_ai(text)
         local prev_pending = ctx.pending_exec_output
         ctx:append_user(text)  -- flushes any pending [exec output] as prefix
-        local resp, err = broker.chat(active_cfg, ctx:to_messages())
-        if not resp then
+
+        local parts = {}
+        local ok, err = broker.chat_stream(active_cfg, ctx:to_messages(),
+            function(delta)
+                parts[#parts + 1] = delta
+                renderer.assistant_delta(delta)
+            end)
+        renderer.assistant_flush()
+
+        if not ok then
             renderer.status("broker error: " .. tostring(err))
             table.remove(ctx.turns)              -- back out the merged user turn
             ctx.pending_exec_output = prev_pending  -- restore buffered exec output
             return
         end
+        local resp = table.concat(parts)
         ctx:append({ role = "assistant", content = resp })
         status_evictions(ctx:enforce_budget())
-        renderer.assistant(resp)
 
         for _, cmd in ipairs(executor.extract_cmd_lines(resp)) do
             local doit