diff --git a/tests/tui_gateway/test_restore_session_usage.py b/tests/tui_gateway/test_restore_session_usage.py
new file mode 100644
index 000000000000..d1a1072c53ef
--- /dev/null
+++ b/tests/tui_gateway/test_restore_session_usage.py
@@ -0,0 +1,160 @@
+"""Tests for _restore_session_usage — token counter restore on session.resume.
+
+When the TUI gateway resumes a session, a fresh agent is built whose
+session_*_tokens counters all start at zero.  _restore_session_usage copies
+the cumulative counts from the stored session row so that session.info /
+session.usage reflects actual usage instead of 0/1.0M-0%.
+"""
+
+import pytest
+
+
+class _FakeAgent:
+    """Minimal stand-in with the same counter attributes as AIAgent."""
+
+    def __init__(self):
+        self.session_input_tokens = 0
+        self.session_output_tokens = 0
+        self.session_cache_read_tokens = 0
+        self.session_cache_write_tokens = 0
+        self.session_reasoning_tokens = 0
+        self.session_total_tokens = 0
+        self.session_api_calls = 0
+        self.session_estimated_cost_usd = 0.0
+        self.session_cost_status = "unknown"
+
+
+def _import_restore():
+    from tui_gateway.server import _restore_session_usage
+    return _restore_session_usage
+
+
+# ── Unit tests for _restore_session_usage ──────────────────────────
+
+
+def test_restore_basic_token_counts():
+    """All five token fields are populated from the stored session row."""
+    restore = _import_restore()
+    agent = _FakeAgent()
+    stored = {
+        "input_tokens": 12000,
+        "output_tokens": 3500,
+        "cache_read_tokens": 800,
+        "cache_write_tokens": 200,
+        "reasoning_tokens": 1500,
+        "api_call_count": 7,
+    }
+    restore(agent, stored)
+
+    assert agent.session_input_tokens == 12000
+    assert agent.session_output_tokens == 3500
+    assert agent.session_cache_read_tokens == 800
+    assert agent.session_cache_write_tokens == 200
+    assert agent.session_reasoning_tokens == 1500
+    assert agent.session_total_tokens == 12000 + 3500 + 800 + 200 + 1500
+    assert agent.session_api_calls == 7
+
+
+def test_restore_total_is_sum():
+    """session_total_tokens equals the sum of all five token components."""
+    restore = _import_restore()
+    agent = _FakeAgent()
+    stored = {
+        "input_tokens": 100,
+        "output_tokens": 200,
+        "cache_read_tokens": 300,
+        "cache_write_tokens": 400,
+        "reasoning_tokens": 500,
+    }
+    restore(agent, stored)
+    assert agent.session_total_tokens == 100 + 200 + 300 + 400 + 500
+
+
+def test_restore_handles_none_values():
+    """Missing or None fields default to zero."""
+    restore = _import_restore()
+    agent = _FakeAgent()
+    stored = {
+        "input_tokens": None,
+        "output_tokens": 50,
+    }
+    restore(agent, stored)
+
+    assert agent.session_input_tokens == 0
+    assert agent.session_output_tokens == 50
+    assert agent.session_cache_read_tokens == 0
+    assert agent.session_cache_write_tokens == 0
+    assert agent.session_reasoning_tokens == 0
+    assert agent.session_total_tokens == 50
+    assert agent.session_api_calls == 0
+
+
+def test_restore_preserves_cost_estimate():
+    """estimated_cost_usd and cost_status are carried over when present."""
+    restore = _import_restore()
+    agent = _FakeAgent()
+    stored = {
+        "input_tokens": 1000,
+        "output_tokens": 500,
+        "estimated_cost_usd": 0.042,
+        "cost_status": "estimated",
+    }
+    restore(agent, stored)
+
+    assert agent.session_estimated_cost_usd == pytest.approx(0.042)
+    assert agent.session_cost_status == "estimated"
+
+
+def test_restore_skips_cost_when_absent():
+    """Cost fields stay at defaults when not in the stored row."""
+    restore = _import_restore()
+    agent = _FakeAgent()
+    stored = {"input_tokens": 100, "output_tokens": 50}
+    restore(agent, stored)
+
+    assert agent.session_estimated_cost_usd == 0.0
+    assert agent.session_cost_status == "unknown"
+
+
+def test_restore_empty_dict():
+    """An empty stored dict leaves the agent at zero counters."""
+    restore = _import_restore()
+    agent = _FakeAgent()
+    restore(agent, {})
+
+    assert agent.session_input_tokens == 0
+    assert agent.session_output_tokens == 0
+    assert agent.session_total_tokens == 0
+    assert agent.session_api_calls == 0
+
+
+# ── _get_usage picks up restored counters ──────────────────────────
+
+
+def test_get_usage_reflects_restored_counters():
+    """After restore, _get_usage returns the stored token counts."""
+    from tui_gateway.server import _get_usage
+
+    agent = _FakeAgent()
+    restore = _import_restore()
+    stored = {
+        "input_tokens": 5000,
+        "output_tokens": 2000,
+        "cache_read_tokens": 100,
+        "cache_write_tokens": 50,
+        "reasoning_tokens": 300,
+        "api_call_count": 3,
+        "estimated_cost_usd": 0.015,
+        "cost_status": "estimated",
+    }
+    restore(agent, stored)
+
+    usage = _get_usage(agent)
+
+    assert usage["input"] == 5000
+    assert usage["output"] == 2000
+    assert usage["cache_read"] == 100
+    assert usage["cache_write"] == 50
+    assert usage["reasoning"] == 300
+    assert usage["total"] == 5000 + 2000 + 100 + 50 + 300
+    assert usage["calls"] == 3
diff --git a/tui_gateway/server.py b/tui_gateway/server.py
index 5df9e51de299..e9cdebebd453 100644
--- a/tui_gateway/server.py
+++ b/tui_gateway/server.py
@@ -1935,6 +1935,38 @@ def _get_usage(agent) -> dict:
     return usage
 
 
+def _restore_session_usage(agent, stored: dict) -> None:
+    """Restore cumulative token counters from a stored session row.
+
+    When ``session.resume`` builds a fresh agent its counters start at zero.
+    The real cumulative usage lives in ``state.db`` (written by
+    ``update_token_counts`` on every API call).  Without this restore the
+    ``session.info`` / ``session.usage`` payloads report all-zero usage
+    which shows as ``0/1.0M-0%`` in the Desktop status bar for gateway
+    sessions viewed after the fact.
+    """
+    agent.session_input_tokens = stored.get("input_tokens") or 0
+    agent.session_output_tokens = stored.get("output_tokens") or 0
+    agent.session_cache_read_tokens = stored.get("cache_read_tokens") or 0
+    agent.session_cache_write_tokens = stored.get("cache_write_tokens") or 0
+    agent.session_reasoning_tokens = stored.get("reasoning_tokens") or 0
+    agent.session_total_tokens = (
+        agent.session_input_tokens
+        + agent.session_output_tokens
+        + agent.session_cache_read_tokens
+        + agent.session_cache_write_tokens
+        + agent.session_reasoning_tokens
+    )
+    agent.session_api_calls = stored.get("api_call_count") or 0
+    # Preserve cost estimate if available
+    cost_usd = stored.get("estimated_cost_usd")
+    if cost_usd:
+        agent.session_estimated_cost_usd = float(cost_usd)
+    cost_status = stored.get("cost_status")
+    if cost_status:
+        agent.session_cost_status = cost_status
+
+
 def _probe_credentials(agent) -> str:
     """Light credential check at session creation — returns warning or ''."""
     try:
@@ -3585,6 +3617,11 @@ def _(rid, params: dict) -> dict:
             # state.db; home override is active here so config/skills/model
             # resolve to the profile too.
             agent = _make_agent(sid, target, session_id=target, session_db=db)
+            # Restore cumulative token counters from the stored session so
+            # that session.info / session.usage shows correct usage instead
+            # of all zeros when viewing gateway sessions in Desktop.
+            if found:
+                _restore_session_usage(agent, found)
         finally:
             _clear_session_context(tokens)
     except Exception as e: