Otto

Automation intelligence for the Frappe ecosystem

Otto is a Frappe Framework app which will be used for adding intelligent automation capabilities to Frappe apps. We're developing it to be a framework for:

1. Automating task handling using.
2. Building custom assistants over Frappe apps.

Otto's app features are built on top of it's library. You may use this in your Frappe app to handle LLM integrations. (Docs)

Otto can currently be used for simple task automation.

If you wish to try it out, here are the following doctypes you can check out:

• Otto Task: definition of a task that can be automated. Task execution is triggered manually or by doctype lifecycle events.
• Otto Tool: definition of a tool that can be used for in a task. Tools run in a server-script like environment.
• Otto LLM: available LLMs that can be used for a task. You can check otto_llm.json for available models.
• Otto Execution: record of a task execution, backed by the purpose agnostic Otto Session doctype.
• Otto Settings: configure your API keys to use the available models.

Most doctype fields have help text associated with them, you may refer to them until proper docs are written.

An internal use case we've applied Otto to is the handling of first response on a support ticket. We're doing this by using our internal Helpdesk Knowledge Base and docs.frappe.io as data sources.

We find that for simpler support queries its performance is limited by the quality and availability of references, and for complex and longer queries it's limited by the model's context coherence. Around 40% of provided feedback is positive (not all handled tickets receive feedback).

This app is part of a larger project in Frappe that aims to explore the best ways to apply whatever state of the art artificial intelligence is available.

Another sub-project is frappe/mcp.

All of these are still in very early stages of development.

Brief example of how the library can be used. Link to full library docs for more details.

import otto.lib as otto
from otto.lib.types import ToolUseUpdate


# 1. Fetch model matching some criteria
model = otto.get_model(supports_vision=True, size="Small", provider="OpenAI") # str model id


# 2. Create new session
session = otto.new(
    model=model,
    instruction="You are a helpful coding assistant",
    tools=[calculator_tool_schema],
)

# Save id to resume session later
session_id = session.id


# 3. Interact (streaming)
stream = session.interact("Calculate 15 * 23", stream=True)
for chunk in stream:
    print(chunk.get("text", ""), end="")
result = stream.item


# 4. Handle tool use
for tool in session.get_pending_tool_use():
    result = execute_tool(tool.name, tool.args) # execute tool

    # update session with tool result
    session.update_tool_use(
        ToolUseUpdate(id=tool.id, status="success", result=result)
    )

# Continue session with tool result
session.interact(stream=False)


# 5. Load and resume interaction (non-streaming)
session = otto.load(session_id)
response, _ = session.interact("What was the result?", stream=False)
if response:
    print(response["content"])

You can install this app using the bench CLI, first setup a Frappe bench directory and create a new site then:

# In you bench directory
bench get-app otto --branch develop
bench --site site-name install-app otto

It is not recommended to use Otto, but should you be curious enough here are some steps you can try out.

1. Set API keys in the Otto Settings doctype.
2. Navigate to Otto LLM select an available model for the key you've provided.
3. Click on the Ask action under Otto LLM to test that the LLM is being called.
4. Navigate to Otto Session to view the previous interaction.
5. Navigate to Otto Task and create a new task:
   • Select a DocType and an event you want to trigger on, for testing you can select ToDo and Manual.
   • Fill out Instruction for what you want your task to do.
   • Fill out Get Context with transformation of the Doc into context info selected.
   • Add tools if require (when creating a new tool, you can check the Mock Tool option)
6. Click on the Execute Task button to test out the task you've created.
7. Click on the Otto Execution link in the dialog and click on Open in Session Viewer (in the right panel)
8. Complete output will only be visible after the task has completed executing.

• Otto Lib Docs