feat(email): voice/style-matched drafting from Sent history (#1607)

docs/guides/email.mdx

@@ -14,6 +14,7 @@ The Email Triage Agent connects to your Gmail account through GAIA's connectors
 - **Organize** — archive, label, mark read/unread, star/unstar. Reversible via the per-action undo log.
 - **Soft-delete with undo** — `trash_message` records the action; `restore_message` reverses it within a 30-second window.
 - **Draft + confirmed send** — generate replies (`draft_reply`) and forwards (`draft_forward`); `send_draft` and `send_now` require explicit user confirmation in the UI.
+- **Voice-matched drafting** — learn your writing style from your Sent mail (`build_voice_profile`) so drafts sound like you; the style profile is derived and stored locally, nothing leaves the device.
 - **Calendar** — list events, accept/decline invites, create events from email content (all calendar mutations gated by user confirmation).
 
 ## Setup
@@ -157,6 +158,10 @@ Senders you reply to quickly are automatically promoted to priority on the next
 
 `profile_inbox` — asks "who emails me most?" and returns a frequency ranking of senders with their dominant category (e.g. *urgent*, *informational*) and the timestamp of their most recent message. Profiling is built from the interaction history the agent accumulates during triage, so it improves the more you use the agent.
 
+### Voice-matched drafting
+
+`build_voice_profile` — ask the agent to "learn my writing style" and it samples your recent Sent mail, derives a style profile (usual greeting, sign-off, typical length, formality), and stores it on-device. From then on, drafted replies come out in your voice instead of neutral boilerplate — still returned for your approval, never auto-sent. The profile keeps derived features only (never your Sent message content) and lives in the agent's local SQLite database; nothing leaves the device. `clear_voice_profile` forgets it.
+
 ### Organize (reversible via the undo log)
 
 `archive_message`, `mark_read`, `mark_unread`, `add_star`, `remove_star`, `label_message`, `move_to_label`

hub/agents/python/email/gaia_agent_email/action_store.py

@@ -68,6 +68,17 @@ class — see ``test_email_agent_soft_delete.py``.
 );
 """
 
+# Voice/style profile derived from Sent mail (#1607). One row per mailbox.
+# ``profile_json`` holds DERIVED features only (greetings, sign-offs, length,
+# formality signals) — never raw Sent content. See ``voice_profile.py``.
+EMAIL_VOICE_PROFILE_DDL = """
+CREATE TABLE IF NOT EXISTS email_voice_profile (
+    mailbox      TEXT PRIMARY KEY,
+    profile_json TEXT NOT NULL,
+    built_at     REAL NOT NULL
+);
+"""
+
 
 # 100 chars max — see plan A4 + adversarial S15. Email bodies routinely
 # carry MFA codes, password reset URLs, banking transaction summaries; a
@@ -76,9 +87,10 @@ class — see ``test_email_agent_soft_delete.py``.
 
 
 def init_schema(db) -> None:
-    """Create both tables if they don't exist, then run migrations. Idempotent."""
+    """Create the tables if they don't exist, then run migrations. Idempotent."""
     db.execute(EMAIL_ACTIONS_DDL)
     db.execute(EMAIL_DRAFTS_DDL)
+    db.execute(EMAIL_VOICE_PROFILE_DDL)
     _migrate_email_actions_mailbox(db)
 
 
@@ -276,16 +288,74 @@ def fetch_draft(db, *, draft_id: str) -> Optional[Dict[str, Any]]:
     return result
 
 
+# ---------------------------------------------------------------------------
+# email_voice_profile API (#1607)
+# ---------------------------------------------------------------------------
+
+
+def save_voice_profile(db, *, mailbox: str, profile: Dict[str, Any]) -> None:
+    """Upsert the voice profile for *mailbox* (one row per mailbox).
+
+    Update-then-insert (not delete-then-insert) so a failure between the
+    two statements can never lose the existing profile.
+    """
+    row = {
+        "profile_json": json.dumps(profile),
+        "built_at": time.time(),
+    }
+    updated = db.update("email_voice_profile", row, "mailbox = :m", {"m": mailbox})
+    if not updated:
+        db.insert("email_voice_profile", dict(row, mailbox=mailbox))
+
+
+def fetch_voice_profile(
+    db, *, mailbox: Optional[str] = None
+) -> Optional[Dict[str, Any]]:
+    """Return the stored profile dict, or ``None`` when none has been built.
+
+    With ``mailbox=None`` returns the most recently built profile across
+    mailboxes — the common single-mailbox case and the system-prompt path.
+    """
+    if mailbox:
+        row = db.query(
+            "SELECT profile_json FROM email_voice_profile WHERE mailbox = :m",
+            {"m": mailbox},
+            one=True,
+        )
+    else:
+        row = db.query(
+            "SELECT profile_json FROM email_voice_profile "
+            "ORDER BY built_at DESC LIMIT 1",
+            {},
+            one=True,
+        )
+    if row is None:
+        return None
+    return json.loads(row["profile_json"])
+
+
+def delete_voice_profile(db, *, mailbox: Optional[str] = None) -> None:
+    """Delete the profile for *mailbox*, or all profiles when ``None``."""
+    if mailbox:
+        db.delete("email_voice_profile", "mailbox = :m", {"m": mailbox})
+    else:
+        db.delete("email_voice_profile", "1 = 1", {})
+
+
 __all__ = [
     "BODY_PREVIEW_MAX_CHARS",
     "EMAIL_ACTIONS_DDL",
     "EMAIL_DRAFTS_DDL",
+    "EMAIL_VOICE_PROFILE_DDL",
+    "delete_voice_profile",
     "fetch_batch_undoable",
     "fetch_draft",
     "fetch_undoable",
+    "fetch_voice_profile",
     "init_schema",
     "mark_draft_sent",
     "mark_undone",
     "record_action",
     "record_draft",
+    "save_voice_profile",
 ]

hub/agents/python/email/gaia_agent_email/agent.py

@@ -60,6 +60,8 @@ class never passes ``use_claude=True`` / ``use_chatgpt=True`` to
 from gaia_agent_email.tools.read_tools import ReadToolsMixin
 from gaia_agent_email.tools.reply_tools import ReplyToolsMixin
 from gaia_agent_email.tools.summarize_tools import SummarizeToolsMixin
+from gaia_agent_email.tools.voice_tools import VoiceToolsMixin
+from gaia_agent_email.voice_profile import render_style_guidance
 
 from gaia.agents.base.agent import Agent
 from gaia.agents.base.console import AgentConsole
@@ -139,6 +141,9 @@ def __getattr__(self, name: str):
   set_category_default, clear_session_preferences) — mutate persistent
   classification preferences that survive across restarts. Confirm the
   change in plain English.
+- Style tools (build_voice_profile, clear_voice_profile) — learn or
+  forget the user's writing style from their Sent mail. Local-only:
+  reads mail, sends nothing; the profile is stored on-device.
 
 PRE-SCAN BEHAVIOR:
 When the user asks for a pre-scan, morning brief, triage view, or "what's
@@ -175,6 +180,7 @@ class EmailTriageAgent(
     PreferenceToolsMixin,
     PhishingToolsMixin,
     ProfileToolsMixin,
+    VoiceToolsMixin,
 ):
     """Email Triage Agent — Gmail + Calendar through the connectors
     framework, all body inference local on Lemonade.
@@ -332,7 +338,13 @@ def _create_console(self) -> AgentConsole:
         return AgentConsole()
 
     def _get_system_prompt(self) -> str:
-        return _SYSTEM_PROMPT
+        # Voice/style-matched drafting (#1607): once a profile has been
+        # built from Sent mail, every turn's prompt carries the style
+        # guidance so draft bodies come out in the user's own voice.
+        profile = action_store.fetch_voice_profile(self)
+        if profile is None:
+            return _SYSTEM_PROMPT
+        return _SYSTEM_PROMPT + "\n" + render_style_guidance(profile)
 
     def process_query(self, *args, **kwargs):
         # Zero the batch-organize counter per turn so a long-lived instance
@@ -357,6 +369,7 @@ def _register_tools(self) -> None:
         self._register_preference_tools()
         self._register_phishing_tools()
         self._register_profile_tools()
+        self._register_voice_tools()
         self.register_memory_tools()
 
     # -- Phase 2 multi-inbox routing (#1603) -------------------------------

hub/agents/python/email/gaia_agent_email/tools/voice_tools.py

@@ -0,0 +1,146 @@
+# Copyright(C) 2025-2026 Advanced Micro Devices, Inc. All rights reserved.
+# SPDX-License-Identifier: MIT
+"""Voice/style profile tools mixin for ``EmailTriageAgent`` (#1607).
+
+``build_voice_profile`` samples the user's Sent mail through the mail
+backend, derives a style profile (``voice_profile.analyze_sent_bodies``),
+and persists it locally via ``action_store``. The agent's system prompt
+picks the stored profile up on the next turn, so draft bodies composed for
+``draft_reply``/``draft_forward`` match the user's own voice.
+
+Both tools are local-only: reading Sent mail mutates nothing remote, and
+the profile lives in the agent's SQLite ``state.db`` — no Sent content
+leaves the device (derived features only are stored).
+"""
+
+from __future__ import annotations
+
+import json
+from typing import Any
+
+from gaia_agent_email import action_store
+from gaia_agent_email.gmail_backend import decode_message_body
+from gaia_agent_email.verbose import log_tool_call
+from gaia_agent_email.voice_profile import analyze_sent_bodies
+
+from gaia.agents.base.tools import tool
+from gaia.connectors.errors import ConnectorsError
+from gaia.connectors.formatting import format_connector_error
+from gaia.logger import get_logger
+
+log = get_logger(__name__)
+
+# Default Sent-mail sample. Big enough to smooth over one-off outliers,
+# small enough to stay fast on first run (one get_message per sample).
+DEFAULT_SAMPLE_SIZE = 50
+
+
+def _envelope_ok(data: Any) -> str:
+    return json.dumps({"ok": True, "data": data}, default=str)
+
+
+def _envelope_err(message: str) -> str:
+    return json.dumps({"ok": False, "error": message})
+
+
+def build_voice_profile_impl(
+    gmail,
+    db,
+    *,
+    mailbox: str,
+    sample_size: int = DEFAULT_SAMPLE_SIZE,
+    debug: bool = False,
+) -> dict:
+    with log_tool_call(
+        "build_voice_profile",
+        {"mailbox": mailbox, "sample_size": sample_size},
+        debug=debug,
+    ) as st:
+        listing = gmail.list_messages(label_ids=["SENT"], max_results=sample_size)
+        refs = listing.get("messages", [])
+        if not refs:
+            raise ValueError(
+                f"no Sent messages found in mailbox '{mailbox}' — a voice "
+                "profile needs Sent history to learn from. Send a few emails "
+                "first, or connect a mailbox that has Sent mail."
+            )
+        bodies = []
+        for ref in refs:
+            msg = gmail.get_message(ref["id"])
+            body, _attachments = decode_message_body(msg.get("payload") or {})
+            if body.strip():
+                bodies.append(body)
+        profile = analyze_sent_bodies(bodies)
+        action_store.save_voice_profile(db, mailbox=mailbox, profile=profile)
+        st["result_summary"] = {"sample_count": profile["sample_count"]}
+        return dict(profile, mailbox=mailbox)
+
+
+class VoiceToolsMixin:
+    """Registers ``build_voice_profile`` and ``clear_voice_profile``."""
+
+    def _register_voice_tools(self) -> None:
+        db = self
+        agent = self
+        debug_flag = bool(getattr(self.config, "debug", False))
+
+        @tool
+        def build_voice_profile(
+            sample_size: int = DEFAULT_SAMPLE_SIZE, mailbox: str = ""
+        ) -> str:
+            """Learn the user's writing voice from their Sent mail.
+
+            Samples recent Sent messages, derives a local style profile
+            (greeting, sign-off, typical length, formality), and stores it
+            on-device. Future draft bodies should match this profile. Reads
+            mail only — sends nothing, mutates nothing remote.
+
+            ``sample_size`` (optional, default 50) caps how many recent Sent
+            messages are analyzed. ``mailbox`` (optional) selects which
+            connected mailbox to learn from; defaults to the primary one.
+            """
+            try:
+                if sample_size <= 0:
+                    return _envelope_err(
+                        f"sample_size must be positive, got {sample_size}"
+                    )
+                if not agent._backends:
+                    return _envelope_err(
+                        "no mailbox is connected — connect Gmail or Outlook "
+                        "via `gaia connectors` first, then retry"
+                    )
+                provider = mailbox or next(iter(agent._backends))
+                backend = agent._backends.get(provider)
+                if backend is None:
+                    return _envelope_err(
+                        f"mailbox '{provider}' is not connected — connected "
+                        f"mailboxes: {sorted(agent._backends)}"
+                    )
+                profile = build_voice_profile_impl(
+                    backend,
+                    db,
+                    mailbox=provider,
+                    sample_size=sample_size,
+                    debug=debug_flag,
+                )
+                return _envelope_ok(profile)
+            except ConnectorsError as exc:
+                return _envelope_err(format_connector_error(exc))
+            except Exception as exc:
+                log.exception("email tool error: %s", type(exc).__name__)
+                return _envelope_err(f"{type(exc).__name__}: {exc}")
+
+        @tool
+        def clear_voice_profile(mailbox: str = "") -> str:
+            """Forget the learned voice profile.
+
+            Removes the stored style profile so drafts go back to neutral
+            phrasing. ``mailbox`` (optional) clears one mailbox's profile;
+            default clears all.
+            """
+            try:
+                action_store.delete_voice_profile(db, mailbox=mailbox or None)
+                return _envelope_ok({"cleared": mailbox or "all"})
+            except Exception as exc:
+                log.exception("email tool error: %s", type(exc).__name__)
+                return _envelope_err(f"{type(exc).__name__}: {exc}")