diff --git a/apps/ai_agent/.github/workflows/ai_agent.yml b/apps/ai_agent/.github/workflows/ai_agent.yml
new file mode 100644
index 00000000..3a6aeab0
--- /dev/null
+++ b/apps/ai_agent/.github/workflows/ai_agent.yml
@@ -0,0 +1,64 @@
+name: AI Agent CI
+
+on:
+  push:
+    paths:
+      - 'apps/ai_agent/**'
+  pull_request:
+    paths:
+      - 'apps/ai_agent/**'
+  schedule:
+    # Weekly Tuesday at 09:00 UTC
+    - cron: '0 9 * * 2'
+
+jobs:
+  test:
+    name: Test
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: apps/ai_agent
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
+
+      - name: Set up Python
+        run: uv python install
+
+      - name: Install dependencies
+        run: uv sync --dev
+
+      - name: Run tests
+        run: uv run pytest
+
+  audit:
+    name: Dependency security audit
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: apps/ai_agent
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
+
+      - name: Set up Python
+        run: uv python install
+
+      - name: Install dependencies (including dev)
+        run: uv sync --dev
+
+      - name: Install pip-audit
+        run: uv add --dev "pip-audit>=2.7"
+
+      - name: Run pip-audit
+        run: uv run pip-audit --requirement <(uv export --no-dev --format requirements-txt)
diff --git a/apps/ai_agent/pyproject.toml b/apps/ai_agent/pyproject.toml
index b1381e96..2b22a74b 100644
--- a/apps/ai_agent/pyproject.toml
+++ b/apps/ai_agent/pyproject.toml
@@ -17,6 +17,7 @@ dev = [
     "httpx>=0.27.0",
     "pytest-mock>=3.14.0",
     "pytest-cov>=5.0.0",
+    "pip-audit>=2.7",
 ]
 
 [dependency-groups]
@@ -24,6 +25,8 @@ dev = [
     "pytest>=8.0.0",
     "pytest-cov>=5.0.0",
     "httpx>=0.27.0",
+    "pytest-mock>=3.14.0",
+    "pip-audit>=2.7",
 ]
 
 [tool.pytest.ini_options]
diff --git a/apps/ai_agent/tests/test_proposals.py b/apps/ai_agent/tests/test_proposals.py
new file mode 100644
index 00000000..3cfe0c32
--- /dev/null
+++ b/apps/ai_agent/tests/test_proposals.py
@@ -0,0 +1,139 @@
+"""Unit tests for POST /proposals/summarise (issue #147)."""
+
+import json
+import pytest
+from unittest.mock import MagicMock, patch
+
+from fastapi.testclient import TestClient
+from main import app
+
+client = TestClient(app)
+
+_BASE_BODY = {
+    "title": "Fund community outreach",
+    "description": "Allocate XLM to grow the Clicked user base across Africa.",
+    "amount": 500.0,
+}
+
+
+def _fake_openai_response(payload: dict):
+    msg = MagicMock()
+    msg.content = json.dumps(payload)
+    choice = MagicMock()
+    choice.message = msg
+    resp = MagicMock()
+    resp.choices = [choice]
+    return resp
+
+
+def _patch_openai(payload: dict):
+    """Context manager: patches _openai_client and returns a configured mock."""
+    patcher = patch("main._openai_client")
+    mock_fn = patcher.start()
+    mock_client = MagicMock()
+    mock_fn.return_value = mock_client
+    mock_client.chat.completions.create.return_value = _fake_openai_response(payload)
+    return patcher, mock_fn
+
+
+def test_happy_path_returns_summary_and_risk():
+    with patch("main._openai_client") as mock_fn:
+        mock_client = MagicMock()
+        mock_fn.return_value = mock_client
+        mock_client.chat.completions.create.return_value = _fake_openai_response(
+            {"summary": "This proposal funds outreach. It is low risk.", "risk": "low"}
+        )
+        response = client.post("/proposals/summarise", json=_BASE_BODY)
+    assert response.status_code == 200
+    data = response.json()
+    assert data["summary"] == "This proposal funds outreach. It is low risk."
+    assert data["risk"] == "low"
+
+
+def test_risk_level_low_accepted():
+    with patch("main._openai_client") as mock_fn:
+        mock_client = MagicMock()
+        mock_fn.return_value = mock_client
+        mock_client.chat.completions.create.return_value = _fake_openai_response(
+            {"summary": "Short summary. Second sentence.", "risk": "low"}
+        )
+        response = client.post("/proposals/summarise", json={**_BASE_BODY, "amount": 10.0})
+    assert response.status_code == 200
+    assert response.json()["risk"] == "low"
+
+
+def test_risk_level_medium_accepted():
+    with patch("main._openai_client") as mock_fn:
+        mock_client = MagicMock()
+        mock_fn.return_value = mock_client
+        mock_client.chat.completions.create.return_value = _fake_openai_response(
+            {"summary": "Moderate proposal. Needs review.", "risk": "medium"}
+        )
+        response = client.post("/proposals/summarise", json=_BASE_BODY)
+    assert response.status_code == 200
+    assert response.json()["risk"] == "medium"
+
+
+def test_risk_level_high_accepted():
+    with patch("main._openai_client") as mock_fn:
+        mock_client = MagicMock()
+        mock_fn.return_value = mock_client
+        mock_client.chat.completions.create.return_value = _fake_openai_response(
+            {"summary": "Very large transfer proposed. High risk detected.", "risk": "high"}
+        )
+        response = client.post("/proposals/summarise", json={**_BASE_BODY, "amount": 1_000_000.0})
+    assert response.status_code == 200
+    assert response.json()["risk"] == "high"
+
+
+def test_empty_summary_returns_502():
+    with patch("main._openai_client") as mock_fn:
+        mock_client = MagicMock()
+        mock_fn.return_value = mock_client
+        mock_client.chat.completions.create.return_value = _fake_openai_response(
+            {"summary": "", "risk": "medium"}
+        )
+        response = client.post("/proposals/summarise", json=_BASE_BODY)
+    assert response.status_code == 502
+
+
+def test_invalid_risk_falls_back_to_medium():
+    with patch("main._openai_client") as mock_fn:
+        mock_client = MagicMock()
+        mock_fn.return_value = mock_client
+        mock_client.chat.completions.create.return_value = _fake_openai_response(
+            {"summary": "Valid summary here. Two sentences total.", "risk": "critical"}
+        )
+        response = client.post("/proposals/summarise", json=_BASE_BODY)
+    assert response.status_code == 200
+    assert response.json()["risk"] == "medium"
+
+
+def test_missing_risk_key_falls_back_to_medium():
+    with patch("main._openai_client") as mock_fn:
+        mock_client = MagicMock()
+        mock_fn.return_value = mock_client
+        mock_client.chat.completions.create.return_value = _fake_openai_response(
+            {"summary": "Summary without risk key. Still valid."}
+        )
+        response = client.post("/proposals/summarise", json=_BASE_BODY)
+    assert response.status_code == 200
+    assert response.json()["risk"] == "medium"
+
+
+def test_missing_api_key_returns_500(monkeypatch):
+    monkeypatch.delenv("OPENAI_API_KEY", raising=False)
+    response = client.post("/proposals/summarise", json=_BASE_BODY)
+    assert response.status_code == 500
+
+
+def test_missing_title_returns_422():
+    body = {k: v for k, v in _BASE_BODY.items() if k != "title"}
+    response = client.post("/proposals/summarise", json=body)
+    assert response.status_code == 422
+
+
+def test_missing_amount_returns_422():
+    body = {k: v for k, v in _BASE_BODY.items() if k != "amount"}
+    response = client.post("/proposals/summarise", json=body)
+    assert response.status_code == 422
diff --git a/apps/ai_agent/tests/test_search.py b/apps/ai_agent/tests/test_search.py
new file mode 100644
index 00000000..911ba2ab
--- /dev/null
+++ b/apps/ai_agent/tests/test_search.py
@@ -0,0 +1,121 @@
+"""Unit tests for GET /search (issue #149)."""
+
+import pytest
+from unittest.mock import MagicMock, patch
+
+from fastapi.testclient import TestClient
+from main import app
+
+client = TestClient(app)
+
+_BASE_PARAMS = {"q": "payment to Alice", "conversationId": "conv-abc"}
+
+
+def _make_weaviate_client(*, exists: bool = True, objects=None):
+    """Build a mock weaviate client."""
+    mock_client = MagicMock()
+    mock_client.collections.exists.return_value = exists
+
+    if objects is not None:
+        mock_result = MagicMock()
+        mock_result.objects = objects
+        mock_client.collections.get.return_value.query.near_vector.return_value = mock_result
+
+    return mock_client
+
+
+def _make_openai_embedding():
+    """Build a mock OpenAI client that returns a dummy embedding."""
+    mock_openai = MagicMock()
+    embed_data = MagicMock()
+    embed_data.embedding = [0.1] * 1536
+    embed_result = MagicMock()
+    embed_result.data = [embed_data]
+    mock_openai.embeddings.create.return_value = embed_result
+    return mock_openai
+
+
+def test_weaviate_connection_failure_returns_503():
+    with patch("main.weaviate.connect_to_local", side_effect=Exception("connection refused")):
+        response = client.get("/search", params=_BASE_PARAMS)
+    assert response.status_code == 503
+
+
+def test_missing_collection_returns_empty_results():
+    """When collection doesn't exist, return empty results without querying Weaviate."""
+    mock_wv = _make_weaviate_client(exists=False)
+    with patch("main.weaviate.connect_to_local", return_value=mock_wv):
+        response = client.get("/search", params=_BASE_PARAMS)
+    assert response.status_code == 200
+    assert response.json() == {"results": []}
+    # Weaviate query must NOT have been called
+    mock_wv.collections.get.return_value.query.near_vector.assert_not_called()
+
+
+def test_returns_results_with_correct_shape():
+    obj = MagicMock()
+    obj.properties = {
+        "messageId": "msg-1",
+        "conversationId": "conv-abc",
+        "senderId": "user-1",
+        "content": "send 50 XLM to Alice",
+    }
+    mock_wv = _make_weaviate_client(exists=True, objects=[obj])
+
+    with patch("main.weaviate.connect_to_local", return_value=mock_wv), \
+         patch("main._openai_client", return_value=_make_openai_embedding()):
+        response = client.get("/search", params=_BASE_PARAMS)
+
+    assert response.status_code == 200
+    data = response.json()
+    assert "results" in data
+    assert len(data["results"]) == 1
+    hit = data["results"][0]
+    assert hit["messageId"] == "msg-1"
+    assert hit["conversationId"] == "conv-abc"
+    assert hit["senderId"] == "user-1"
+    assert hit["content"] == "send 50 XLM to Alice"
+
+
+def test_filters_by_conversation_id():
+    obj = MagicMock()
+    obj.properties = {
+        "messageId": "msg-2",
+        "conversationId": "conv-xyz",
+        "senderId": "user-2",
+        "content": "transfer 100 XLM",
+    }
+    mock_wv = _make_weaviate_client(exists=True, objects=[obj])
+
+    with patch("main.weaviate.connect_to_local", return_value=mock_wv), \
+         patch("main._openai_client", return_value=_make_openai_embedding()):
+        response = client.get("/search", params={"q": "transfer", "conversationId": "conv-xyz"})
+
+    assert response.status_code == 200
+    # Verify the near_vector call was made (filter is passed inside it)
+    mock_wv.collections.get.return_value.query.near_vector.assert_called_once()
+    call_kwargs = mock_wv.collections.get.return_value.query.near_vector.call_args[1]
+    # The filter argument must be present
+    assert "filters" in call_kwargs
+
+
+def test_close_called_on_success():
+    mock_wv = _make_weaviate_client(exists=True, objects=[])
+    mock_wv.collections.get.return_value.query.near_vector.return_value.objects = []
+
+    with patch("main.weaviate.connect_to_local", return_value=mock_wv), \
+         patch("main._openai_client", return_value=_make_openai_embedding()):
+        response = client.get("/search", params=_BASE_PARAMS)
+
+    assert response.status_code == 200
+    mock_wv.close.assert_called_once()
+
+
+def test_missing_q_returns_422():
+    response = client.get("/search", params={"conversationId": "conv-abc"})
+    assert response.status_code == 422
+
+
+def test_missing_conversation_id_returns_422():
+    response = client.get("/search", params={"q": "hello"})
+    assert response.status_code == 422
diff --git a/apps/ai_agent/tests/test_transfers.py b/apps/ai_agent/tests/test_transfers.py
index e3216ec8..dc1deb28 100644
--- a/apps/ai_agent/tests/test_transfers.py
+++ b/apps/ai_agent/tests/test_transfers.py
@@ -87,3 +87,72 @@ def test_llm_path_missing_api_key_returns_500(monkeypatch):
         "amount": 100.0, "sender": "GABC", "recipient": "GDEF", "memo": "test"
     })
     assert response.status_code == 500
+
+
+# ── Rule-based path (issue #145) ──────────────────────────────────────────────
+
+def test_high_value_transfer_is_flagged_without_llm_call():
+    """High-value path bypasses the LLM — mandatory security property."""
+    with patch("main._openai_client") as mock_openai:
+        response = client.post(
+            "/transfers/analyse",
+            json={
+                "amount": 10_001.0,
+                "sender": "GABC",
+                "recipient": "GDEF",
+                "memo": "large payment",
+            },
+        )
+        assert response.status_code == 200
+        data = response.json()
+        assert data["flagged"] is True
+        assert data["confidence"] == 0.99
+        mock_openai.assert_not_called()
+
+
+def test_high_value_reason_mentions_threshold():
+    """The flagging reason must reference the threshold value."""
+    with patch("main._openai_client"):
+        response = client.post(
+            "/transfers/analyse",
+            json={
+                "amount": 50_000.0,
+                "sender": "GABC",
+                "recipient": "GDEF",
+                "memo": "bulk transfer",
+            },
+        )
+        assert response.status_code == 200
+        data = response.json()
+        assert data["flagged"] is True
+        reason = data["reason"] or ""
+        assert "10000" in reason or "threshold" in reason.lower()
+
+
+def test_amount_exactly_at_threshold_takes_llm_path():
+    """amount == threshold is NOT above it; should route to LLM."""
+    with patch("main._openai_client") as mock_openai_fn:
+        mock_client = MagicMock()
+        mock_openai_fn.return_value = mock_client
+        mock_client.chat.completions.create.return_value = _fake_openai_response(
+            {"flagged": False, "reason": None, "confidence": 0.1}
+        )
+        response = client.post(
+            "/transfers/analyse",
+            json={
+                "amount": 10_000.0,
+                "sender": "GABC",
+                "recipient": "GDEF",
+                "memo": "exactly at threshold",
+            },
+        )
+        assert response.status_code == 200
+        mock_openai_fn.assert_called_once()
+
+
+def test_missing_fields_return_422():
+    response = client.post(
+        "/transfers/analyse",
+        json={"sender": "GABC", "recipient": "GDEF", "memo": "no amount"},
+    )
+    assert response.status_code == 422