diff --git a/CLAUDE.md b/CLAUDE.md
index 4fc24fb..21a279e 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -553,7 +553,13 @@ Beyond the three reverted changes, the following hardening is live (see git hist
 
 - [ ] Real-time calendar availability check (`odoo/calendar.py`)
 - [ ] Whisper large-v3 post-call transcription (`recording/transcriber.py`)
-- [ ] Recording + transcript attached to Odoo lead chatter
+- [x] Recording + transcript attached to Odoo lead chatter (2026-07-04, pulled forward):
+      `odoo_client.attach_call_artifacts` posts the transcript as a chatter note + .txt
+      attachment and attaches the stereo WAV; wired post-call in `bot.py` (both the
+      extraction path and the in-call tool path stamp `artifacts["odoo_ref"]`). Transcript
+      is the live in-call Whisper `medium` text (`extract.format_transcript`), not
+      large-v3 — the large-v3 re-transcription remains a Phase 3 item. Verified e2e
+      against db1 (create → attach → verify → delete, left clean).
 - [ ] Staff review flow confirmed in Odoo
 
 **Gate:** Staff receives, reviews, and confirms a lead end-to-end
diff --git a/bot.py b/bot.py
index bcf724c..45ebae4 100644
--- a/bot.py
+++ b/bot.py
@@ -650,12 +650,17 @@ async def run_agent(transport, caller_number=None, call_sid=None, do_capture=Tru
     # can stamp the verified caller-ID + call_sid onto the lead (the model never supplies a
     # phone number — we don't ask for one). With tools on, this writes the Odoo lead IN-CALL,
     # so the post-call extraction is skipped below to avoid a duplicate.
+    # Where the call's artifacts land, filled in as the call runs: the saved recording
+    # path (audiobuffer handler) and the Odoo ref of the created lead ("odoo:<model>:<id>"),
+    # so the WAV + transcript can be attached to the lead after the call.
+    artifacts = {"recording": None, "odoo_ref": None}
+
     if ENABLE_TOOLS:
         async def _record_appointment(params):
             args = params.arguments or {}
             if do_capture:
                 from practice import persist_appointment
-                persist_appointment({
+                artifacts["odoo_ref"] = persist_appointment({
                     "call_sid": call_sid,
                     "patient_name": args.get("patient_name"),
                     "callback_number": caller_number,  # verified caller-ID, not model-supplied
@@ -807,6 +812,7 @@ async def run_agent(transport, caller_number=None, call_sid=None, do_capture=Tru
                 wf.setsampwidth(2)
                 wf.setframerate(sample_rate)
                 wf.writeframes(audio)
+            artifacts["recording"] = path
             logger.info(f"Saved call recording: {path} "
                         f"({len(audio)} bytes, {num_channels}ch @ {sample_rate}Hz)")
 
@@ -840,13 +846,42 @@ async def run_agent(transport, caller_number=None, call_sid=None, do_capture=Tru
         try:
             from extract import extract_and_record
 
-            await extract_and_record(
+            rec = await extract_and_record(
                 context.messages, OLLAMA_URL, OLLAMA_MODEL,
                 call_sid=call_sid, caller_number=caller_number,
             )
+            if rec:
+                artifacts["odoo_ref"] = rec.get("persisted_to")
         except Exception:
             logger.exception("Post-call appointment extraction failed")
 
+    # Attach the call recording + transcript to the Odoo lead (Phase 3: staff review a
+    # lead with the audio and transcript right on it). Only when a lead actually landed
+    # in Odoo; JSONL-fallback leads get artifacts when re-imported... they stay on disk.
+    if do_capture and (artifacts["odoo_ref"] or "").startswith("odoo:"):
+        try:
+            from extract import format_transcript
+            from odoo_client import attach_call_artifacts
+
+            _, odoo_model, odoo_id = artifacts["odoo_ref"].split(":")
+            # The recording is written by an async handler after stop_recording —
+            # give it a moment if it hasn't landed yet.
+            for _ in range(10):
+                if not RECORD_CALLS or artifacts["recording"]:
+                    break
+                await asyncio.sleep(0.5)
+            await asyncio.to_thread(
+                attach_call_artifacts,
+                odoo_model, int(odoo_id),
+                wav_path=artifacts["recording"],
+                transcript=format_transcript(context.messages, agent_name=AGENT_NAME),
+                call_sid=call_sid,
+            )
+            logger.info(f"Attached recording+transcript to {artifacts['odoo_ref']} "
+                        f"(wav={artifacts['recording']})")
+        except Exception:
+            logger.exception("Attaching call artifacts to Odoo failed (lead itself is safe)")
+
 
 async def run_call(websocket, serializer: TwilioFrameSerializer, caller_number=None, call_sid=None):
     """Phone entrypoint: wrap the Twilio Media Stream in a transport, run the shared agent."""
diff --git a/extract.py b/extract.py
index 878319a..7befe96 100644
--- a/extract.py
+++ b/extract.py
@@ -42,6 +42,20 @@ _EXTRACT_INSTRUCTIONS = (
 )
 
 
+def format_transcript(messages, agent_name="AVA") -> str:
+    """Plain-text transcript of the call (Caller/agent turns) for chatter/attachments."""
+    lines = []
+    for m in messages:
+        if m.get("role") not in ("user", "assistant"):
+            continue
+        content = m.get("content")
+        if not isinstance(content, str) or not content.strip():
+            continue
+        who = "Caller" if m["role"] == "user" else agent_name
+        lines.append(f"{who}: {content.strip()}")
+    return "\n".join(lines)
+
+
 async def extract_and_record(messages, ollama_url, model, call_sid=None, caller_number=None):
     """Extract an appointment from the transcript and persist it. Returns the record
     dict if one was saved, else None."""
@@ -136,6 +150,7 @@ async def extract_and_record(messages, ollama_url, model, call_sid=None, caller_
         "source": "post_call_extraction",
     }
     where = persist_appointment(record)
+    record["persisted_to"] = where  # "odoo:<model>:<id>" | "jsonl" — used to attach artifacts
     logger.info(
         f"Post-call {kind} saved ({where}): {record['patient_name']} / "
         f"{record['location']} / reason={record['reason']} / ins={record['insurance']} / "
diff --git a/odoo_client.py b/odoo_client.py
index 02bf605..1ca9c62 100644
--- a/odoo_client.py
+++ b/odoo_client.py
@@ -145,3 +145,40 @@ def create_appointment_request(patient_name, callback_number, reason, preferred_
         vals["user_id"] = ODOO_USER_ID
     rec = _exec(uid, models, "crm.lead", "create", vals)
     return ("crm.lead", rec)
+
+
+def attach_call_artifacts(model, rec_id, wav_path=None, transcript=None, call_sid=None):
+    """Attach the call recording (WAV) and post the transcript to a lead's chatter.
+
+    Called post-call for the record created by create_appointment_request. The transcript
+    goes in as a chatter note (readable in place) AND as a .txt attachment (downloadable);
+    the WAV is attached as-is. Any failure raises — the caller logs and moves on (the lead
+    itself is already safe)."""
+    import base64
+
+    uid, models = _connect()
+    sid = call_sid or "call"
+
+    if transcript:
+        body = "<b>📼 Call transcript</b><br/>" + "<br/>".join(
+            escape(line) for line in transcript.splitlines() if line.strip()
+        )
+        _exec(uid, models, model, "message_post", [rec_id], body=body)
+        _exec(uid, models, "ir.attachment", "create", [{
+            "name": f"transcript_{sid}.txt",
+            "datas": base64.b64encode(transcript.encode()).decode(),
+            "res_model": model,
+            "res_id": rec_id,
+            "mimetype": "text/plain",
+        }])
+
+    if wav_path and os.path.isfile(wav_path):
+        with open(wav_path, "rb") as fh:
+            b64 = base64.b64encode(fh.read()).decode()
+        _exec(uid, models, "ir.attachment", "create", [{
+            "name": os.path.basename(wav_path),
+            "datas": b64,
+            "res_model": model,
+            "res_id": rec_id,
+            "mimetype": "audio/wav",
+        }])