diff --git a/llm/cli.py b/llm/cli.py
index 2e11e2c8..89a46e24 100644
--- a/llm/cli.py
+++ b/llm/cli.py
@@ -1014,6 +1014,14 @@ async def inner():
     default=5,
     help="How many chained tool responses to allow, default 5, set 0 for unlimited",
 )
+@click.option(
+    "-n",
+    "--count",
+    type=int,
+    is_flag=False,
+    flag_value=3,
+    help="Number of previous messages to show. Requires -c or --cid. Default 3, set 0 for unlimited",
+)
 def chat(
     system,
     model_id,
@@ -1032,6 +1040,7 @@ def chat(
     tools_debug,
     tools_approve,
     chain_limit,
+    count,
 ):
     """
     Hold an ongoing chat with a model.
@@ -1146,6 +1155,28 @@ def chat(
         raise click.ClickException(str(ex))
 
     click.echo("Chatting with {}".format(model.model_id))
+
+    if conversation and conversation.responses and count is not None:
+        if count <= 0:
+            history_label = "All Messages"
+            start_index = 0
+        else:
+            history_label = "Last {} Messages".format(count)
+            # Calculate the start index to get the last 'count' turns
+            start_index = max(0, len(conversation.responses) - count)
+
+        click.echo("\n--- Conversation History ({}) ---".format(history_label))
+
+
+        # Iterate over the relevant slice of conversation responses
+        for i, response_obj in enumerate(conversation.responses[start_index:]):
+            # Each response_obj is an llm.Response object, representing a turn
+            click.echo(f"\n# {response_obj.datetime_utc()}  id: {response_obj.id}")
+            click.echo(f"\n\n## Prompt\n{response_obj.prompt.prompt}")
+            click.echo(f"\n\n## Response\n{response_obj}")
+
+        click.echo("\n--- End History ---")
+
     click.echo("Type 'exit' or 'quit' to exit")
     click.echo("Type '!multi' to enter multiple lines, then '!end' to finish")
     click.echo("Type '!edit' to open your default editor and modify the prompt")
diff --git a/llm/models.py b/llm/models.py
index 9aa4801f..75e688b3 100644
--- a/llm/models.py
+++ b/llm/models.py
@@ -765,6 +765,7 @@ def from_row(cls, db, row, _async=False):
         )
         prompt_json = json.loads(row["prompt_json"] or "null")
         response.id = row["id"]
+        response._start_utcnow = datetime.datetime.fromisoformat(row["datetime_utc"])
         response._prompt_json = prompt_json
         response.response_json = json.loads(row["response_json"] or "null")
         response._done = True
@@ -1141,7 +1142,9 @@ def duration_ms(self) -> int:
 
     def datetime_utc(self) -> str:
         self._force()
-        return self._start_utcnow.isoformat() if self._start_utcnow else ""
+        if self._start_utcnow:
+            return self._start_utcnow.strftime("%Y-%m-%dT%H:%M:%S")
+        return ""
 
     def usage(self) -> Usage:
         self._force()
diff --git a/tests/test_chat_history.py b/tests/test_chat_history.py
new file mode 100644
index 00000000..2edf2edf
--- /dev/null
+++ b/tests/test_chat_history.py
@@ -0,0 +1,330 @@
+# tests/test_chat_history.py
+import pytest
+import sys
+import textwrap
+from click.testing import CliRunner
+from unittest.mock import ANY
+import llm.cli
+import sqlite_utils
+import datetime
+from ulid import ULID
+
+
+# Fixture to pre-populate the logs database with a conversation
+@pytest.fixture
+def populated_logs_db(user_path):
+    """
+    Returns a logs_db instance pre-populated with a conversation
+    containing 5 message turns, located at user_path/logs.db.
+    """
+    log_db_path = user_path / "logs.db"
+    db = sqlite_utils.Database(str(log_db_path))
+
+    # Ensure migrations are run for the database
+    llm.cli.migrate(db)
+
+    # Create a consistent conversation ID for these test messages
+    conversation_id = str(ULID()).lower()
+
+    # Insert conversation record
+    db["conversations"].insert({
+        "id": conversation_id,
+        "name": "Pre-populated Chat History",
+        "model": "mock",
+    }, pk="id")
+
+    prompts = [f"Prompt {i}" for i in range(1, 6)]
+    responses_text = [f"Response {i}" for i in range(1, 6)]
+
+    generated_datetimes = []
+    generated_ulids = []
+
+    # Populate responses table with a history of 5 turns
+    # Ensure timestamps are unique and chronologically ordered for consistent history loading
+    for i in range(5):
+        timestamp = datetime.datetime.now(datetime.timezone.utc) + datetime.timedelta(seconds=i)
+        response_ulid = str(ULID.from_datetime(timestamp)).lower()
+
+        # Store the datetime object for later use in tests
+        generated_datetimes.append(timestamp)
+        generated_ulids.append(response_ulid);
+
+        db["responses"].insert({
+            "id": response_ulid,
+            "model": "mock",
+            "resolved_model": None,
+            "prompt": prompts[i],
+            "system": None,
+            "prompt_json": None,
+            "options_json": "{}",
+            "response": responses_text[i],
+            "response_json": None,
+            "conversation_id": conversation_id,
+            "duration_ms": 100,
+            "datetime_utc": timestamp.isoformat(timespec='microseconds'),
+            "input_tokens": len(prompts[i].split()),
+            "output_tokens": len(responses_text[i].split()),
+            "token_details": None,
+            "schema_id": None,
+        }, pk="id")
+
+    # Return the db instance, conversation_id, AND the generated datetimes
+    return db, conversation_id, generated_datetimes, generated_ulids
+
+@pytest.mark.xfail(sys.platform == "win32", reason="Expected to fail on Windows")
+def test_chat_continue_with_bare_l(mock_model, populated_logs_db):
+    runner = CliRunner()
+    # Unpack all three values from the fixture
+    db, conversation_id, generated_datetimes, generated_ulids = populated_logs_db
+
+    mock_model.enqueue(["New Response"]) # Response for the new prompt
+
+    result = runner.invoke(
+        llm.cli.cli,
+        ["chat", "-m", "mock", "-c", "--count"], # --count defaults to 3 messages
+        input="New Prompt\nquit\n",
+        catch_exceptions=False,
+    )
+
+    assert result.exit_code == 0
+    output_lines = result.output.splitlines()
+
+    # Define the formatting function that matches your datetime_utc() method
+    def format_datetime_for_output(dt_obj):
+        if dt_obj:
+            # This must match the format you use in Response.datetime_utc()
+            return dt_obj.strftime("%Y-%m-%dT%H:%M:%S")
+        return ""
+
+    # Expected formatted datetimes and ULIDs for the last 3 messages (indices 2, 3, 4)
+    expected_dt3 = format_datetime_for_output(generated_datetimes[2])
+    expected_id3 = generated_ulids[2] # Use the stored ULID string
+    expected_dt4 = format_datetime_for_output(generated_datetimes[3])
+    expected_id4 = generated_ulids[3] # Use the stored ULID string
+    expected_dt5 = format_datetime_for_output(generated_datetimes[4])
+    expected_id5 = generated_ulids[4] # Use the stored ULID string
+
+    # Check for expected headings
+    assert "--- Conversation History (Last 3 Messages) ---" in output_lines
+    assert "## Prompt" in output_lines
+    assert "## Response" in output_lines
+    assert "--- End History ---" in output_lines
+
+    # Update assertions to use both formatted date and stored ULID
+    assert f"# {expected_dt3}  id: {expected_id3}" in output_lines
+    assert "Prompt 3" in output_lines
+    assert "Response 3" in output_lines
+
+    assert f"# {expected_dt4}  id: {expected_id4}" in output_lines
+    assert "Prompt 4" in output_lines
+    assert "Response 4" in output_lines
+
+    assert f"# {expected_dt5}  id: {expected_id5}" in output_lines
+    assert "Prompt 5" in output_lines
+    assert "Response 5" in output_lines
+
+    # Ensure older messages are NOT in history
+    assert "Prompt 1" not in output_lines
+    assert "Prompt 2" not in output_lines
+
+    # Ensure older messages are NOT in history
+    assert "Prompt 1" not in output_lines
+    assert "Prompt 2" not in output_lines
+
+    # Verify order
+    index_prompt3_line = output_lines.index("Prompt 3")
+    index_prompt4_line = output_lines.index("Prompt 4")
+    index_prompt5_line = output_lines.index("Prompt 5")
+
+    assert index_prompt3_line < index_prompt4_line < index_prompt5_line
+
+    assert "> New Prompt" in output_lines
+    assert "New Response" in output_lines
+
+
+@pytest.mark.xfail(sys.platform == "win32", reason="Expected to fail on Windows")
+def test_chat_continue_with_l_set_to_1(mock_model, populated_logs_db):
+    runner = CliRunner()
+    # Unpack all three values from the fixture
+    db, conversation_id, *_ = populated_logs_db
+
+    mock_model.enqueue(["New Response"]) # Response for the new prompt
+
+    result = runner.invoke(
+        llm.cli.cli,
+        ["chat", "-m", "mock", "-c", "-n", "1"],
+        input="New Prompt\nquit\n",
+        catch_exceptions=False,
+    )
+
+    assert result.exit_code == 0
+    output_lines = result.output.splitlines()
+
+    assert "--- Conversation History (Last 1 Messages) ---" in output_lines
+    # Only the last turn should be in history
+    assert "## Prompt" in output_lines
+    assert "Prompt 5" in output_lines
+    assert "## Response" in output_lines
+    assert "Response 5" in output_lines
+
+    # Ensure older messages are NOT in history
+    assert "Prompt 4" not in output_lines
+    assert "Prompt 3" not in output_lines
+    assert "Prompt 2" not in output_lines
+    assert "Prompt 1" not in output_lines
+
+    assert "--- End History ---" in output_lines
+
+    assert "> New Prompt" in output_lines
+    assert "New Response" in output_lines
+
+
+@pytest.mark.xfail(sys.platform == "win32", reason="Expected to fail on Windows")
+def test_chat_continue_with_l_set_to_2(mock_model, populated_logs_db):
+    runner = CliRunner()
+    # Unpack all three values from the fixture
+    db, conversation_id, *_ = populated_logs_db
+
+    mock_model.enqueue(["New Response"]) # Response for the new prompt
+
+    result = runner.invoke(
+        llm.cli.cli,
+        ["chat", "-m", "mock", "-c", "-n", "2"],
+        input="New Prompt\nquit\n",
+        catch_exceptions=False,
+    )
+
+    assert result.exit_code == 0
+    output_lines = result.output.splitlines()
+
+    # Check for history header
+    assert "--- Conversation History (Last 2 Messages) ---" in output_lines
+
+    # Check that the last two turns are displayed
+    assert "## Prompt" in output_lines # Check for the label itself
+    assert "Prompt 4" in output_lines
+    assert "## Response" in output_lines # Check for the label itself
+    assert "Response 4" in output_lines
+    assert "Prompt 5" in output_lines
+    assert "Response 5" in output_lines
+
+    # Ensure older messages are NOT in history
+    assert "Prompt 1" not in output_lines
+    assert "Prompt 2" not in output_lines
+    assert "Prompt 3" not in output_lines
+
+    # Check for history footer
+    assert "--- End History ---" in output_lines
+
+    # Ensure the new prompt and response are also present
+    assert "> New Prompt" in output_lines
+    assert "New Response" in output_lines
+
+    # Verify order: Prompt 4, Response 4, Prompt 5, Response 5
+    # Find indices of the specific content lines
+    index_prompt4_line = output_lines.index("Prompt 4")
+    index_response4_line = output_lines.index("Response 4")
+    index_prompt5_line = output_lines.index("Prompt 5")
+    index_response5_line = output_lines.index("Response 5")
+
+    assert index_prompt4_line < index_response4_line < index_prompt5_line < index_response5_line
+
+@pytest.mark.xfail(sys.platform == "win32", reason="Expected to fail on Windows")
+def test_chat_continue_with_l_set_to_0(mock_model, populated_logs_db):
+    runner = CliRunner()
+    # Unpack all three values from the fixture
+    db, conversation_id, *_ = populated_logs_db
+
+    mock_model.enqueue(["New Response"]) # Response for the new prompt
+
+    result = runner.invoke(
+        llm.cli.cli,
+        ["chat", "-m", "mock", "-c", "-n", "0"],
+        input="New Prompt\nquit\n",
+        catch_exceptions=False,
+    )
+
+    assert result.exit_code == 0
+    output_lines = result.output.splitlines()
+
+    # Check for history header
+    assert "--- Conversation History (All Messages) ---" in output_lines
+
+    # Check that all turns are displayed
+    assert "## Prompt" in output_lines # Check for the label itself
+    assert "## Response" in output_lines # Check for the label itself
+    assert "Prompt 1" in output_lines
+    assert "Response 1" in output_lines
+    assert "Prompt 2" in output_lines
+    assert "Response 3" in output_lines
+    assert "Prompt 3" in output_lines
+    assert "Response 4" in output_lines
+    assert "Prompt 5" in output_lines
+    assert "Response 5" in output_lines
+
+    # Check for history footer
+    assert "--- End History ---" in output_lines
+
+    # Ensure the new prompt and response are also present
+    assert "> New Prompt" in output_lines
+    assert "New Response" in output_lines
+
+    # Verify order: Prompt 1, Response 1, Prompt 5, Response 5
+    # Find indices of the specific content lines
+    index_prompt1_line = output_lines.index("Prompt 1")
+    index_response1_line = output_lines.index("Response 1")
+    index_prompt5_line = output_lines.index("Prompt 5")
+    index_response5_line = output_lines.index("Response 5")
+
+    assert index_prompt1_line < index_response1_line < index_prompt5_line < index_response5_line
+
+
+@pytest.mark.xfail(sys.platform == "win32", reason="Expected to fail on Windows")
+def test_chat_l_without_c_is_ignored(mock_model, populated_logs_db):
+    runner = CliRunner()
+    db, original_conversation_id, *_ = populated_logs_db # Get the pre-populated db and conversation_id
+
+    mock_model.enqueue(["New Chat Response"]) # Response for the new prompt
+
+    result = runner.invoke(
+        llm.cli.cli,
+        ["chat", "-m", "mock", "-n", "1"], # -l specified, but -c omitted
+        input="New Prompt\nquit\n",
+        catch_exceptions=False,
+    )
+
+    assert result.exit_code == 0
+    output_lines = result.output.splitlines()
+
+    # History header should NOT be present because -c was not used
+    assert "--- Conversation History" not in result.output
+    assert "--- End History ---" not in result.output
+
+    # The new prompt and response should be present
+    assert "> New Prompt" in output_lines
+    assert "New Chat Response" in output_lines
+
+    # Old conversation messages from the pre-populated DB should not be displayed
+    assert "Prompt 5" not in output_lines
+    assert "Response 5" not in output_lines
+
+    # Verify that a new conversation was started in the logs
+    # We expect 2 conversations: the original populated one, and the new one.
+    conversations = list(db["conversations"].rows)
+    assert len(conversations) == 2
+
+    # Get the ID of the newly created conversation
+    new_conversation_id = None
+    for convo in conversations:
+        if convo["id"] != original_conversation_id:
+            new_conversation_id = convo["id"]
+            break
+    assert new_conversation_id is not None
+
+    # Verify the last logged response belongs to the new conversation
+    responses = list(db["responses"].rows)
+    assert responses[-1]["prompt"] == "New Prompt"
+    assert responses[-1]["conversation_id"] == new_conversation_id
+    assert responses[-1]["conversation_id"] != original_conversation_id
+
+