Add back history and reset subcommand in magics (jupyterlab#997)

* Add back history and reset subcommand in magics

This time it's not specific to OpenAI.

* Document `%ai reset`

* Add the `max_history` magic command option

* Respect `max_history` when calling the model

All history is still kept so to allow configuring a higher `max_history`
mid-session and have older exchanges be included.

* Use `<role>:` prefix instead of pseudo-XML for non-chat models

* Document the `max_history` setting for the magic commands

* fix: make `max_history` a configuration option

* fix: `max_history=0` to generate with no context

* Add unit test for `max_history` magic option

* Add unit test for `%ai reset` magic command

* Test that transcript is updated with new prompt

* Review corrections

- `context` instead of `self.transcript` as if clause condition
- less confusing model context input variable

docs/source/users/index.md

@@ -785,6 +785,34 @@ A function that computes the lowest common multiples of two integers, and
 a function that runs 5 test cases of the lowest common multiple function
 ```
 
+### Configuring the amount of history to include in the context
+
+By default, two previous Human/AI message exchanges are included in the context of the new prompt.
+You can change this using the IPython `%config` magic, for example:
+
+```python
+%config AiMagics.max_history = 4
+```
+
+Note that old messages are still kept locally in memory,
+so they will be included in the context of the next prompt after raising the `max_history` value.
+
+You can configure the value for all notebooks
+by specifying `c.AiMagics.max_history` traitlet in `ipython_config.py`, for example:
+
+```python
+c.AiMagics.max_history = 4
+```
+
+### Clearing the chat history
+
+You can run the `%ai reset` line magic command to clear the chat history. After you do this,
+previous magic commands you've run will no longer be added as context in requests.
+
+```
+%ai reset
+```
+
 ### Interpolating in prompts
 
 Using curly brace syntax, you can include variables and other Python expressions in your

packages/jupyter-ai-magics/jupyter_ai_magics/magics.py

@@ -15,6 +15,7 @@
 from jupyter_ai_magics.utils import decompose_model_id, get_lm_providers
 from langchain.chains import LLMChain
 from langchain.schema import HumanMessage
+from langchain_core.messages import AIMessage
 
 from ._version import __version__
 from .parsers import (
@@ -24,6 +25,7 @@
     HelpArgs,
     ListArgs,
     RegisterArgs,
+    ResetArgs,
     UpdateArgs,
     VersionArgs,
     cell_magic_parser,
@@ -144,9 +146,18 @@ class AiMagics(Magics):
         config=True,
     )
 
+    max_history = traitlets.Int(
+        default_value=2,
+        allow_none=False,
+        help="""Maximum number of exchanges (user/assistant) to include in the history
+        when invoking a chat model, defaults to 2.
+        """,
+        config=True,
+    )
+
     def __init__(self, shell):
         super().__init__(shell)
-        self.transcript_openai = []
+        self.transcript = []
 
         # suppress warning when using old Anthropic provider
         warnings.filterwarnings(
@@ -437,6 +448,12 @@ def handle_error(self, args: ErrorArgs):
 
         return self.run_ai_cell(cell_args, prompt)
 
+    def _append_exchange(self, prompt: str, output: str):
+        """Appends a conversational exchange between user and an OpenAI Chat
+        model to a transcript that will be included in future exchanges."""
+        self.transcript.append(HumanMessage(prompt))
+        self.transcript.append(AIMessage(output))
+
     def _decompose_model_id(self, model_id: str):
         """Breaks down a model ID into a two-tuple (provider_id, local_model_id). Returns (None, None) if indeterminate."""
         # custom_model_registry maps keys to either a model name (a string) or an LLMChain.
@@ -500,6 +517,9 @@ def handle_list(self, args: ListArgs):
     def handle_version(self, args: VersionArgs):
         return __version__
 
+    def handle_reset(self, args: ResetArgs):
+        self.transcript = []
+
     def run_ai_cell(self, args: CellArgs, prompt: str):
         provider_id, local_model_id = self._decompose_model_id(args.model_id)
 
@@ -577,13 +597,29 @@ def run_ai_cell(self, args: CellArgs, prompt: str):
         ip = self.shell
         prompt = prompt.format_map(FormatDict(ip.user_ns))
 
+        context = self.transcript[-2 * self.max_history :] if self.max_history else []
         if provider.is_chat_provider:
-            result = provider.generate([[HumanMessage(content=prompt)]])
+            result = provider.generate([[*context, HumanMessage(content=prompt)]])
         else:
             # generate output from model via provider
-            result = provider.generate([prompt])
+            if context:
+                inputs = [
+                    (
+                        f"AI: {message.content}"
+                        if message.type == "ai"
+                        else f"{message.type.title()}: {message.content}"
+                    )
+                    for message in context + [HumanMessage(content=prompt)]
+                ]
+            else:
+                inputs = [prompt]
+            result = provider.generate(inputs)
 
         output = result.generations[0][0].text
+
+        # append exchange to transcript
+        self._append_exchange(prompt, output)
+
         md = {"jupyter_ai": {"provider_id": provider_id, "model_id": local_model_id}}
 
         return self.display_output(output, args.format, md)
@@ -628,6 +664,8 @@ def ai(self, line, cell=None):
                 return self.handle_update(args)
             if args.type == "version":
                 return self.handle_version(args)
+            if args.type == "reset":
+                return self.handle_reset(args)
         except ValueError as e:
             print(e, file=sys.stderr)
             return

packages/jupyter-ai-magics/jupyter_ai_magics/parsers.py

@@ -95,6 +95,10 @@ class UpdateArgs(BaseModel):
     target: str
 
 
+class ResetArgs(BaseModel):
+    type: Literal["reset"] = "reset"
+
+
 class LineMagicGroup(click.Group):
     """Helper class to print the help string for cell magics as well when
     `%ai --help` is called."""
@@ -277,3 +281,12 @@ def register_subparser(**kwargs):
 def register_subparser(**kwargs):
     """Update an alias called NAME to refer to the model or chain named TARGET."""
     return UpdateArgs(**kwargs)
+
+
+@line_magic_parser.command(
+    name="reset",
+    short_help="Clear the conversation transcript.",
+)
+def register_subparser(**kwargs):
+    """Clear the conversation transcript."""
+    return ResetArgs()

packages/jupyter-ai-magics/jupyter_ai_magics/tests/test_magics.py

@@ -1,7 +1,11 @@
-from unittest.mock import patch
+import os
+from unittest.mock import Mock, patch
 
+import pytest
 from IPython import InteractiveShell
+from IPython.core.display import Markdown
 from jupyter_ai_magics.magics import AiMagics
+from langchain_core.messages import AIMessage, HumanMessage
 from pytest import fixture
 from traitlets.config.loader import Config
 
@@ -48,3 +52,71 @@ def test_default_model_error_line(ip):
         assert mock_run.called
         cell_args = mock_run.call_args.args[0]
         assert cell_args.model_id == "my-favourite-llm"
+
+
+PROMPT = HumanMessage(
+    content=("Write code for me please\n\nProduce output in markdown format only.")
+)
+RESPONSE = AIMessage(content="Leet code")
+AI1 = AIMessage("ai1")
+H1 = HumanMessage("h1")
+AI2 = AIMessage("ai2")
+H2 = HumanMessage("h2")
+AI3 = AIMessage("ai3")
+
+
+@pytest.mark.parametrize(
+    ["transcript", "max_history", "expected_context"],
+    [
+        ([], 3, [PROMPT]),
+        ([AI1], 0, [PROMPT]),
+        ([AI1], 1, [AI1, PROMPT]),
+        ([H1, AI1], 0, [PROMPT]),
+        ([H1, AI1], 1, [H1, AI1, PROMPT]),
+        ([AI1, H1, AI2], 0, [PROMPT]),
+        ([AI1, H1, AI2], 1, [H1, AI2, PROMPT]),
+        ([AI1, H1, AI2], 2, [AI1, H1, AI2, PROMPT]),
+        ([H1, AI1, H2, AI2], 0, [PROMPT]),
+        ([H1, AI1, H2, AI2], 1, [H2, AI2, PROMPT]),
+        ([H1, AI1, H2, AI2], 2, [H1, AI1, H2, AI2, PROMPT]),
+        ([AI1, H1, AI2, H2, AI3], 0, [PROMPT]),
+        ([AI1, H1, AI2, H2, AI3], 1, [H2, AI3, PROMPT]),
+        ([AI1, H1, AI2, H2, AI3], 2, [H1, AI2, H2, AI3, PROMPT]),
+        ([AI1, H1, AI2, H2, AI3], 3, [AI1, H1, AI2, H2, AI3, PROMPT]),
+    ],
+)
+def test_max_history(ip, transcript, max_history, expected_context):
+    ip.extension_manager.load_extension("jupyter_ai_magics")
+    ai_magics = ip.magics_manager.registry["AiMagics"]
+    ai_magics.transcript = transcript.copy()
+    ai_magics.max_history = max_history
+    provider = ai_magics._get_provider("openrouter")
+    with patch.object(provider, "generate") as generate, patch.dict(
+        os.environ, OPENROUTER_API_KEY="123"
+    ):
+        generate.return_value.generations = [[Mock(text="Leet code")]]
+        result = ip.run_cell_magic(
+            "ai",
+            "openrouter:anthropic/claude-3.5-sonnet",
+            cell="Write code for me please",
+        )
+        provider.generate.assert_called_once_with([expected_context])
+    assert isinstance(result, Markdown)
+    assert result.data == "Leet code"
+    assert result.filename is None
+    assert result.metadata == {
+        "jupyter_ai": {
+            "model_id": "anthropic/claude-3.5-sonnet",
+            "provider_id": "openrouter",
+        }
+    }
+    assert result.url is None
+    assert ai_magics.transcript == [*transcript, PROMPT, RESPONSE]
+
+
+def test_reset(ip):
+    ip.extension_manager.load_extension("jupyter_ai_magics")
+    ai_magics = ip.magics_manager.registry["AiMagics"]
+    ai_magics.transcript = [AI1, H1, AI2, H2, AI3]
+    result = ip.run_line_magic("ai", "reset")
+    assert ai_magics.transcript == []