diff --git a/README.md b/README.md
index caf7e5a..ec96304 100644
--- a/README.md
+++ b/README.md
@@ -132,7 +132,8 @@ let cfg = Config {
     listener_pool:      2,                        // Supervised accept-loop actors
     scheduler_threads:  Some(2),                  // smarm worker threads (None = one per CPU)
     keep_alive_timeout: Duration::from_secs(60),  // Idle budget between requests
-    request_timeout:    Duration::from_secs(30),  // Whole-request read deadline
+    request_timeout:    Duration::from_secs(30),  // Whole-request READ deadline
+    write_timeout:      Duration::from_secs(30),  // Per-write response budget
     max_header_count:   64,
     read_buf_size:      8 * 1024,
     max_body_bytes:     1 << 20,
@@ -152,7 +153,14 @@ serve_with(cfg, pipeline).unwrap();
   until its head and body are fully read. Expiry mid-head gets a
   best-effort `408 Request Timeout`; expiry mid-body just closes. Deadlines
   are absolute instants, so a trickling (slowloris-style) client can't
-  reset its budget by sending one byte at a time.
+  reset its budget by sending one byte at a time. Covers the **read phase
+  only** — a streaming response may legitimately outlive any whole-request
+  clock.
+- `write_timeout` — a per-write budget for response bytes: the fixed
+  head+body write, and *each* streamed chunk, must complete within it. A
+  client that stops reading is dropped once its socket buffer fills and a
+  write stalls past the budget (the write-side twin of slowloris). Streamed
+  chunks each get a fresh budget; a stream as a whole has no deadline.
 
 ### Graceful Shutdown
 
@@ -185,6 +193,62 @@ If every `Handle` is dropped, shutdown can never be signalled and the
 server runs forever — exactly `serve_with`'s semantics (it does this
 internally).
 
+### Streaming Responses
+
+A handler can return a body it doesn't have yet: hand urus the read end of
+a channel and feed it from a producer actor (the conn actor owns the
+socket and all write deadlines; the producer never touches the fd):
+
+```rust
+use urus::{Conn, Next};
+
+|c: Conn, _n: Next| {
+    let (tx, rx) = smarm::channel::<Vec<u8>>();
+    smarm::spawn(move || {
+        for part in ["chunk one", "chunk two"] {
+            if tx.send(part.as_bytes().to_vec()).is_err() {
+                return; // connection died — stop producing
+            }
+        }
+        // tx drops: end of stream.
+    });
+    c.put_status(200).put_body(rx)
+}
+```
+
+On HTTP/1.1 the response uses `Transfer-Encoding: chunked` and the
+connection is reusable afterwards; on HTTP/1.0 (no chunked framing) bytes
+go out raw and the connection closes to delimit the body. End of stream is
+signalled by dropping every `Sender`. The producer contract: a `send`
+error means the connection is gone — exit. Chunked **request** bodies are
+decoded transparently; handlers see `conn.body` either way.
+
+### Server-Sent Events
+
+`Conn::sse()` is the streaming body with the SSE headers and a heartbeat
+wired up:
+
+```rust
+|c: Conn, _n: Next| {
+    let (c, events) = c.sse(); // or c.sse_with_heartbeat(interval)
+    smarm::spawn(move || {
+        loop {
+            if events.send("tick", "data").is_err() {
+                return; // SseClosed: client gone or server draining
+            }
+            smarm::sleep(std::time::Duration::from_secs(1));
+        }
+    });
+    c
+}
+```
+
+While the producer is silent the connection actor emits `: keep-alive`
+comment chunks every `HEARTBEAT_INTERVAL` (15s default). Liveness comes
+from that ping: a vanished client is detected when a write stalls past
+`write_timeout`. No request clock applies to an open SSE stream, and a
+graceful shutdown force-stops it at the drain deadline.
+
 ### Named Actors
 
 For introspection and debugging, the server registers itself in smarm's
@@ -262,9 +326,9 @@ This avoids the complexity of async ecosystems while maintaining full concurrenc
 ## Roadmap
 
 See [`ROADMAP.md`](ROADMAP.md). Summary: v0.2 (supervised listener pool,
-graceful shutdown, enforced timeouts, registry names) is done; next up are
-streaming bodies + SSE (v0.3), WebSocket (v0.4), PubSub (v0.5), and
-Phoenix-style channels (v0.6).
+graceful shutdown, enforced timeouts, registry names) and v0.3 (streaming
+bodies, chunked requests, SSE) are done; next up are WebSocket (v0.4),
+PubSub (v0.5), and Phoenix-style channels (v0.6).
 
 Refer to `urus-spec.md` and `urus-v1-build-notes.md` in the artifact persistence for the original design and implementation notes.
 
diff --git a/ROADMAP.md b/ROADMAP.md
index 7ffe8c4..0ff45ac 100644
--- a/ROADMAP.md
+++ b/ROADMAP.md
@@ -119,22 +119,43 @@ panicking listener restarts under load without dropped accepts (test each).
 
 ---
 
-## v0.3 — Streaming bodies + SSE
+## v0.3 — Streaming bodies + SSE ✅ (landed 2026-06, chunks 1-3)
 
 The slipped v1 goal, and the WebSocket prerequisite (an upgraded connection
 is "a response that never ends").
 
-1. `RespBody::Stream(...)` — handler hands back a writer callback or a
-   `Receiver<Vec<u8>>`; conn actor switches to chunked transfer-encoding
-   (HTTP/1.1) and pumps until the stream closes. Keep-alive after a
-   completed chunked response.
-2. Chunked **request** bodies (`parser.rs` currently returns 411) — same
-   cycle, the codec knowledge is fresh.
-3. SSE sugar: `Conn::sse()` → an `EventSender` (`.send(event, data)`),
-   sets `content-type: text/event-stream`, disables buffering, heartbeats
-   via comment lines on a timer. An SSE conn is just a Stream body with
-   the per-read deadline disabled (or set to the heartbeat interval) —
-   liveness comes from the keep-alive ping, not `request_timeout`.
+1. `RespBody::Stream(StreamBody)` ✅ — **decided: pull** (user call at the
+   v0.3 design fork): the handler hands back a `smarm::Receiver<Vec<u8>>`
+   and the producer is an actor the handler spawned; the conn actor pumps
+   (`pump_stream`) and keeps sole ownership of the socket and write
+   deadlines. The recv() park between chunks is stoppable by the draining
+   registry, so infinite streams die at the drain deadline with no
+   polling. HTTP/1.1: chunked TE, keep-alive after the terminator;
+   HTTP/1.0: raw bytes, forced close, EOF-delimited. End of stream =
+   every Sender dropped = 0-chunk. Timeout split (handoff Q3, as
+   proposed): `request_timeout` covers the READ phase only; new
+   `Config.write_timeout` (default 30s) bounds every response write —
+   fresh budget per streamed chunk, and the fixed-body + error-response
+   writes went deadline-bounded too (write-side slowloris closed).
+2. Chunked **request** bodies ✅ — `read_chunked_body` decodes
+   incrementally on the request deadline; extensions ignored, trailers
+   consumed + discarded, decoded size capped at max_body_bytes (413),
+   size-line/trailer spam capped (400). CL + TE: chunked, or chunked on
+   HTTP/1.0, rejected 400 at parse (smuggling ambiguity). Keep-alive
+   drain accounts RAW framing length, so pipelined requests behind a
+   chunked body land exactly.
+3. SSE sugar ✅ — `Conn::sse()` / `sse_with_heartbeat` → `EventSender`
+   (`send(event, data)` / `data` / `comment`, Clone, `Err(SseClosed)`
+   once the stream is gone). Heartbeat lives in the pump as leaned: with
+   `StreamBody.heartbeat` set, the chunk wait is `recv_timeout(interval)`
+   and expiry writes a `: keep-alive` comment chunk. Liveness from the
+   ping (dead client = heartbeat write stalls past write_timeout), not
+   from any request clock.
+
+**Exit criteria (verified):** chunked round-trips both directions with
+keep-alive and pipelining intact; an infinite SSE stream heartbeats while
+silent, is killed by write_timeout when the client stops reading, and is
+force-stopped at the drain deadline on shutdown; full suite green.
 
 ---
 
diff --git a/examples/crud.rs b/examples/crud.rs
index 02cedea..55de4ae 100644
--- a/examples/crud.rs
+++ b/examples/crud.rs
@@ -204,6 +204,28 @@ fn parse_id(s: &str) -> Option<u64> {
 // Handlers
 // ---------------------------------------------------------------------------
 
+/// SSE demo: `curl -N localhost:8080/ticker` streams a tick every second
+/// (with `: keep-alive` comments if it ever goes quiet). The producer
+/// exits on SseClosed (client gone / write timeout / shutdown drain) or
+/// when the example is shutting down.
+fn ticker(conn: Conn, _next: Next) -> Conn {
+    let (conn, events) = conn.sse();
+    smarm::spawn(move || {
+        let mut n: u64 = 0;
+        loop {
+            if SHUTTING_DOWN.load(std::sync::atomic::Ordering::Relaxed) {
+                return; // dropping `events` ends the stream cleanly
+            }
+            if events.send("tick", &n.to_string()).is_err() {
+                return; // SseClosed
+            }
+            n += 1;
+            smarm::sleep(std::time::Duration::from_secs(1));
+        }
+    });
+    conn
+}
+
 fn list(conn: Conn, _next: Next) -> Conn {
     let (tx, rx) = channel::<(u16, Vec<u8>)>();
     store().send(Request::List { reply: tx }).ok();
@@ -278,7 +300,8 @@ fn main() {
                 .post(  "/users",     create)
                 .get(   "/users/:id", get_one)
                 .put(   "/users/:id", update)
-                .delete("/users/:id", delete),
+                .delete("/users/:id", delete)
+                .get(   "/ticker",    ticker),
         );
 
     let cfg = Config::new("127.0.0.1:8080".parse().unwrap());