discourse-ai/lib/ai_bot/playground.rb
Sam a7d032fa28
DEV: artifact system update (#1096)
### Why

This pull request fundamentally restructures how AI bots create and update web artifacts to address critical limitations in the previous approach:

1.  **Improved Artifact Context for LLMs**: Previously, artifact creation and update tools included the *entire* artifact source code directly in the tool arguments. This overloaded the Language Model (LLM) with raw code, making it difficult for the LLM to maintain a clear understanding of the artifact's current state when applying changes. The LLM would struggle to differentiate between the base artifact and the requested modifications, leading to confusion and less effective updates.
2.  **Reduced Token Usage and History Bloat**: Including the full artifact source code in every tool interaction was extremely token-inefficient.  As conversations progressed, this redundant code in the history consumed a significant number of tokens unnecessarily. This not only increased costs but also diluted the context for the LLM with less relevant historical information.
3.  **Enabling Updates for Large Artifacts**: The lack of a practical diff or targeted update mechanism made it nearly impossible to efficiently update larger web artifacts.  Sending the entire source code for every minor change was both computationally expensive and prone to errors, effectively blocking the use of AI bots for meaningful modifications of complex artifacts.

**This pull request addresses these core issues by**:

*   Introducing methods for the AI bot to explicitly *read* and understand the current state of an artifact.
*   Implementing efficient update strategies that send *targeted* changes rather than the entire artifact source code.
*   Providing options to control the level of artifact context included in LLM prompts, optimizing token usage.

### What

The main changes implemented in this PR to resolve the above issues are:

1.  **`Read Artifact` Tool for Contextual Awareness**:
    - A new `read_artifact` tool is introduced, enabling AI bots to fetch and process the current content of a web artifact from a given URL (local or external).
    - This provides the LLM with a clear and up-to-date representation of the artifact's HTML, CSS, and JavaScript, improving its understanding of the base to be modified.
    - By cloning local artifacts, it allows the bot to work with a fresh copy, further enhancing context and control.

2.  **Refactored `Update Artifact` Tool with Efficient Strategies**:
    - The `update_artifact` tool is redesigned to employ more efficient update strategies, minimizing token usage and improving update precision:
        - **`diff` strategy**:  Utilizes a search-and-replace diff algorithm to apply only the necessary, targeted changes to the artifact's code. This significantly reduces the amount of code sent to the LLM and focuses its attention on the specific modifications.
        - **`full` strategy**:  Provides the option to replace the entire content sections (HTML, CSS, JavaScript) when a complete rewrite is required.
    - Tool options enhance the control over the update process:
        - `editor_llm`:  Allows selection of a specific LLM for artifact updates, potentially optimizing for code editing tasks.
        - `update_algorithm`: Enables choosing between `diff` and `full` update strategies based on the nature of the required changes.
        - `do_not_echo_artifact`:  Defaults to true, and by *not* echoing the artifact in prompts, it further reduces token consumption in scenarios where the LLM might not need the full artifact context for every update step (though effectiveness might be slightly reduced in certain update scenarios).

3.  **System and General Persona Tool Option Visibility and Customization**:
    - Tool options, including those for system personas, are made visible and editable in the admin UI. This allows administrators to fine-tune the behavior of all personas and their tools, including setting specific LLMs or update algorithms. This was previously limited or hidden for system personas.

4.  **Centralized and Improved Content Security Policy (CSP) Management**:
    - The CSP for AI artifacts is consolidated and made more maintainable through the `ALLOWED_CDN_SOURCES` constant. This improves code organization and future updates to the allowed CDN list, while maintaining the existing security posture.

5.  **Codebase Improvements**:
    - Refactoring of diff utilities, introduction of strategy classes, enhanced error handling, new locales, and comprehensive testing all contribute to a more robust, efficient, and maintainable artifact management system.

By addressing the issues of LLM context confusion, token inefficiency, and the limitations of updating large artifacts, this pull request significantly improves the practicality and effectiveness of AI bots in managing web artifacts within Discourse.
2025-02-04 16:27:27 +11:00

640 lines
21 KiB
Ruby

# frozen_string_literal: true
module DiscourseAi
module AiBot
class Playground
BYPASS_AI_REPLY_CUSTOM_FIELD = "discourse_ai_bypass_ai_reply"
attr_reader :bot
# An abstraction to manage the bot and topic interactions.
# The bot will take care of completions while this class updates the topic title
# and stream replies.
def self.find_chat_persona(message, channel, user)
if channel.direct_message_channel?
AiPersona
.allowed_modalities(allow_chat_direct_messages: true)
.find do |p|
p[:user_id].in?(channel.allowed_user_ids) && (user.group_ids & p[:allowed_group_ids])
end
else
# let's defer on the parse if there is no @ in the message
if message.message.include?("@")
mentions = message.parsed_mentions.parsed_direct_mentions
if mentions.present?
AiPersona
.allowed_modalities(allow_chat_channel_mentions: true)
.find { |p| p[:username].in?(mentions) && (user.group_ids & p[:allowed_group_ids]) }
end
end
end
end
def self.schedule_chat_reply(message, channel, user, context)
return if !SiteSetting.ai_bot_enabled
all_chat =
AiPersona.allowed_modalities(
allow_chat_channel_mentions: true,
allow_chat_direct_messages: true,
)
return if all_chat.blank?
return if all_chat.any? { |m| m[:user_id] == user.id }
persona = find_chat_persona(message, channel, user)
return if !persona
post_ids = nil
post_ids = context.dig(:context, :post_ids) if context.is_a?(Hash)
::Jobs.enqueue(
:create_ai_chat_reply,
channel_id: channel.id,
message_id: message.id,
persona_id: persona[:id],
context_post_ids: post_ids,
)
end
def self.is_bot_user_id?(user_id)
# this will catch everything and avoid any feedback loops
# we could get feedback loops between say discobot and ai-bot or third party plugins
# and bots
user_id.to_i <= 0
end
def self.schedule_reply(post)
return if is_bot_user_id?(post.user_id)
mentionables = nil
if post.topic.private_message?
mentionables =
AiPersona.allowed_modalities(user: post.user, allow_personal_messages: true)
else
mentionables = AiPersona.allowed_modalities(user: post.user, allow_topic_mentions: true)
end
bot_user = nil
mentioned = nil
all_llm_users =
LlmModel
.where(enabled_chat_bot: true)
.joins(:user)
.pluck("users.id", "users.username_lower")
if post.topic.private_message?
# this is an edge case, you started a PM with a different bot
bot_user =
post.topic.topic_allowed_users.where(user_id: all_llm_users.map(&:first)).first&.user
bot_user ||=
post
.topic
.topic_allowed_users
.where(user_id: mentionables.map { |m| m[:user_id] })
.first
&.user
end
mentions = nil
if mentionables.present? || (bot_user && post.topic.private_message?)
mentions = post.mentions.map(&:downcase)
# in case we are replying to a post by a bot
if post.reply_to_post_number && post.reply_to_post&.user
mentions << post.reply_to_post.user.username_lower
end
end
if mentionables.present?
mentioned = mentionables.find { |mentionable| mentions.include?(mentionable[:username]) }
# direct PM to mentionable
if !mentioned && bot_user
mentioned = mentionables.find { |mentionable| bot_user.id == mentionable[:user_id] }
end
# public topic so we need to use the persona user
bot_user ||= User.find_by(id: mentioned[:user_id]) if mentioned
end
if bot_user && post.reply_to_post_number && !post.reply_to_post.user&.bot?
# replying to a non-bot user
return
end
if bot_user
topic_persona_id = post.topic.custom_fields["ai_persona_id"]
persona_id = mentioned&.dig(:id) || topic_persona_id
persona = nil
if persona_id
persona =
DiscourseAi::AiBot::Personas::Persona.find_by(user: post.user, id: persona_id.to_i)
end
if !persona && persona_name = post.topic.custom_fields["ai_persona"]
persona =
DiscourseAi::AiBot::Personas::Persona.find_by(user: post.user, name: persona_name)
end
# edge case, llm was mentioned in an ai persona conversation
if persona_id == topic_persona_id.to_i && post.topic.private_message? && persona &&
all_llm_users.present?
if !persona.force_default_llm && mentions.present?
mentioned_llm_user_id, _ =
all_llm_users.find { |id, username| mentions.include?(username) }
if mentioned_llm_user_id
bot_user = User.find_by(id: mentioned_llm_user_id) || bot_user
end
end
end
persona ||= DiscourseAi::AiBot::Personas::General
bot_user = User.find(persona.user_id) if persona && persona.force_default_llm
bot = DiscourseAi::AiBot::Bot.as(bot_user, persona: persona.new)
new(bot).update_playground_with(post)
end
end
def initialize(bot)
@bot = bot
end
def update_playground_with(post)
schedule_bot_reply(post) if can_attach?(post)
end
def conversation_context(post)
# Pay attention to the `post_number <= ?` here.
# We want to inject the last post as context because they are translated differently.
# also setting default to 40, allowing huge contexts costs lots of tokens
max_posts = 40
if bot.persona.class.respond_to?(:max_context_posts)
max_posts = bot.persona.class.max_context_posts || 40
end
post_types = [Post.types[:regular]]
post_types << Post.types[:whisper] if post.post_type == Post.types[:whisper]
context =
post
.topic
.posts
.joins(:user)
.joins("LEFT JOIN post_custom_prompts ON post_custom_prompts.post_id = posts.id")
.where("post_number <= ?", post.post_number)
.order("post_number desc")
.where("post_type in (?)", post_types)
.limit(max_posts)
.pluck(
"posts.raw",
"users.username",
"post_custom_prompts.custom_prompt",
"(
SELECT array_agg(ref.upload_id)
FROM upload_references ref
WHERE ref.target_type = 'Post' AND ref.target_id = posts.id
) as upload_ids",
)
builder = DiscourseAi::Completions::PromptMessagesBuilder.new
context.reverse_each do |raw, username, custom_prompt, upload_ids|
custom_prompt_translation =
Proc.new do |message|
# We can't keep backwards-compatibility for stored functions.
# Tool syntax requires a tool_call_id which we don't have.
if message[2] != "function"
custom_context = {
content: message[0],
type: message[2].present? ? message[2].to_sym : :model,
}
custom_context[:id] = message[1] if custom_context[:type] != :model
custom_context[:name] = message[3] if message[3]
builder.push(**custom_context)
end
end
if custom_prompt.present?
custom_prompt.each(&custom_prompt_translation)
else
context = {
content: raw,
type: (available_bot_usernames.include?(username) ? :model : :user),
}
context[:id] = username if context[:type] == :user
if upload_ids.present? && context[:type] == :user && bot.persona.class.vision_enabled
context[:upload_ids] = upload_ids.compact
end
builder.push(**context)
end
end
builder.to_a
end
def title_playground(post, user)
context = conversation_context(post)
bot
.get_updated_title(context, post, user)
.tap do |new_title|
PostRevisor.new(post.topic.first_post, post.topic).revise!(
bot.bot_user,
title: new_title.sub(/\A"/, "").sub(/"\Z/, ""),
)
end
allowed_users = post.topic.topic_allowed_users.pluck(:user_id)
MessageBus.publish(
"/discourse-ai/ai-bot/topic/#{post.topic.id}",
{ title: post.topic.title },
user_ids: allowed_users,
)
end
def chat_context(message, channel, persona_user, context_post_ids)
has_vision = bot.persona.class.vision_enabled
include_thread_titles = !channel.direct_message_channel? && !message.thread_id
current_id = message.id
if !channel.direct_message_channel?
# we are interacting via mentions ... strip mention
instruction_message = message.message.gsub(/@#{bot.bot_user.username}/i, "").strip
end
messages = nil
max_messages = 40
if bot.persona.class.respond_to?(:max_context_posts)
max_messages = bot.persona.class.max_context_posts || 40
end
if !message.thread_id && channel.direct_message_channel?
messages = [message]
elsif !channel.direct_message_channel? && !message.thread_id
messages =
Chat::Message
.joins("left join chat_threads on chat_threads.id = chat_messages.thread_id")
.where(chat_channel_id: channel.id)
.where(
"chat_messages.thread_id IS NULL OR chat_threads.original_message_id = chat_messages.id",
)
.order(id: :desc)
.limit(max_messages)
.to_a
.reverse
end
messages ||=
ChatSDK::Thread.last_messages(
thread_id: message.thread_id,
guardian: Discourse.system_user.guardian,
page_size: max_messages,
)
builder = DiscourseAi::Completions::PromptMessagesBuilder.new
guardian = Guardian.new(message.user)
if context_post_ids
builder.set_chat_context_posts(context_post_ids, guardian, include_uploads: has_vision)
end
messages.each do |m|
# restore stripped message
m.message = instruction_message if m.id == current_id && instruction_message
if available_bot_user_ids.include?(m.user_id)
builder.push(type: :model, content: m.message)
else
upload_ids = nil
upload_ids = m.uploads.map(&:id) if has_vision && m.uploads.present?
mapped_message = m.message
thread_title = nil
thread_title = m.thread&.title if include_thread_titles && m.thread_id
mapped_message = "(#{thread_title})\n#{m.message}" if thread_title
builder.push(
type: :user,
content: mapped_message,
name: m.user.username,
upload_ids: upload_ids,
)
end
end
builder.to_a(
limit: max_messages,
style: channel.direct_message_channel? ? :chat_with_context : :chat,
)
end
def reply_to_chat_message(message, channel, context_post_ids)
persona_user = User.find(bot.persona.class.user_id)
participants = channel.user_chat_channel_memberships.map { |m| m.user.username }
context_post_ids = nil if !channel.direct_message_channel?
context =
get_context(
participants: participants.join(", "),
conversation_context: chat_context(message, channel, persona_user, context_post_ids),
user: message.user,
skip_tool_details: true,
)
reply = nil
guardian = Guardian.new(persona_user)
force_thread = message.thread_id.nil? && channel.direct_message_channel?
in_reply_to_id = channel.direct_message_channel? ? message.id : nil
new_prompts =
bot.reply(context) do |partial, cancel, placeholder|
if !reply
# just eat all leading spaces we can not create the message
next if partial.blank?
reply =
ChatSDK::Message.create(
raw: partial,
thread_id: message.thread_id,
channel_id: channel.id,
guardian: guardian,
in_reply_to_id: in_reply_to_id,
force_thread: force_thread,
enforce_membership: !channel.direct_message_channel?,
)
ChatSDK::Message.start_stream(message_id: reply.id, guardian: guardian)
else
streaming =
ChatSDK::Message.stream(message_id: reply.id, raw: partial, guardian: guardian)
if !streaming
cancel&.call
break
end
end
end
if new_prompts.length > 1 && reply.id
ChatMessageCustomPrompt.create!(message_id: reply.id, custom_prompt: new_prompts)
end
ChatSDK::Message.stop_stream(message_id: reply.id, guardian: guardian) if reply
reply
end
def get_context(participants:, conversation_context:, user:, skip_tool_details: nil)
result = {
site_url: Discourse.base_url,
site_title: SiteSetting.title,
site_description: SiteSetting.site_description,
time: Time.zone.now,
participants: participants,
conversation_context: conversation_context,
user: user,
}
result[:skip_tool_details] = true if skip_tool_details
result
end
def reply_to(post, custom_instructions: nil, &blk)
# this is a multithreading issue
# post custom prompt is needed and it may not
# be properly loaded, ensure it is loaded
PostCustomPrompt.none
reply = +""
post_streamer = nil
post_type =
post.post_type == Post.types[:whisper] ? Post.types[:whisper] : Post.types[:regular]
context =
get_context(
participants: post.topic.allowed_users.map(&:username).join(", "),
conversation_context: conversation_context(post),
user: post.user,
)
context[:post_id] = post.id
context[:topic_id] = post.topic_id
context[:private_message] = post.topic.private_message?
context[:custom_instructions] = custom_instructions
reply_user = bot.bot_user
if bot.persona.class.respond_to?(:user_id)
reply_user = User.find_by(id: bot.persona.class.user_id) || reply_user
end
stream_reply = post.topic.private_message?
# we need to ensure persona user is allowed to reply to the pm
if post.topic.private_message?
if !post.topic.topic_allowed_users.exists?(user_id: reply_user.id)
post.topic.topic_allowed_users.create!(user_id: reply_user.id)
end
end
if stream_reply
reply_post =
PostCreator.create!(
reply_user,
topic_id: post.topic_id,
raw: "",
skip_validations: true,
skip_jobs: true,
post_type: post_type,
)
publish_update(reply_post, { raw: reply_post.cooked })
redis_stream_key = "gpt_cancel:#{reply_post.id}"
Discourse.redis.setex(redis_stream_key, 60, 1)
end
context[:skip_tool_details] ||= !bot.persona.class.tool_details
post_streamer = PostStreamer.new(delay: Rails.env.test? ? 0 : 0.5) if stream_reply
new_custom_prompts =
bot.reply(context) do |partial, cancel, placeholder, type|
reply << partial
raw = reply.dup
raw << "\n\n" << placeholder if placeholder.present?
if blk && type != :tool_details && type != :partial_tool && type != :partial_invoke
blk.call(partial)
end
if stream_reply && !Discourse.redis.get(redis_stream_key)
cancel&.call
reply_post.update!(raw: reply, cooked: PrettyText.cook(reply))
# we do not break out, cause if we do
# we will not get results from bot
# leading to broken context
# we need to trust it to cancel at the endpoint
end
if post_streamer
post_streamer.run_later do
Discourse.redis.expire(redis_stream_key, 60)
publish_update(reply_post, { raw: raw })
end
end
end
return if reply.blank?
if stream_reply
post_streamer.finish
post_streamer = nil
# land the final message prior to saving so we don't clash
reply_post.cooked = PrettyText.cook(reply)
publish_final_update(reply_post)
reply_post.revise(
bot.bot_user,
{ raw: reply },
skip_validations: true,
skip_revision: true,
)
else
reply_post =
PostCreator.create!(
reply_user,
topic_id: post.topic_id,
raw: reply,
skip_validations: true,
post_type: post_type,
)
end
# we do not need to add a custom prompt for a single reply
if new_custom_prompts.length > 1
reply_post.post_custom_prompt ||= reply_post.build_post_custom_prompt(custom_prompt: [])
prompt = reply_post.post_custom_prompt.custom_prompt || []
prompt.concat(new_custom_prompts)
reply_post.post_custom_prompt.update!(custom_prompt: prompt)
end
reply_post
rescue => e
if reply_post
details = e.message.to_s
reply = "#{reply}\n\n#{I18n.t("discourse_ai.ai_bot.reply_error", details: details)}"
reply_post.revise(
bot.bot_user,
{ raw: reply },
skip_validations: true,
skip_revision: true,
)
end
raise e
ensure
# since we are skipping validations and jobs we
# may need to fix participant count
if reply_post && reply_post.topic && reply_post.topic.private_message? &&
reply_post.topic.participant_count < 2
reply_post.topic.update!(participant_count: 2)
end
post_streamer&.finish(skip_callback: true)
publish_final_update(reply_post) if stream_reply
if reply_post && post.post_number == 1 && post.topic.private_message?
title_playground(reply_post, post.user)
end
end
def available_bot_usernames
@bot_usernames ||=
AiPersona.joins(:user).pluck(:username).concat(available_bot_users.map(&:username))
end
def available_bot_user_ids
@bot_ids ||= AiPersona.joins(:user).pluck("users.id").concat(available_bot_users.map(&:id))
end
private
def available_bot_users
@available_bots ||=
User.joins("INNER JOIN llm_models llm ON llm.user_id = users.id").where(active: true)
end
def publish_final_update(reply_post)
return if @published_final_update
if reply_post
publish_update(reply_post, { cooked: reply_post.cooked, done: true })
# we subscribe at position -2 so we will always get this message
# moving all cooked on every page load is wasteful ... this means
# we have a benign message at the end, 2 is set to ensure last message
# is delivered
publish_update(reply_post, { noop: true })
@published_final_update = true
end
end
def can_attach?(post)
return false if bot.bot_user.nil?
return false if post.topic.private_message? && post.post_type != Post.types[:regular]
return false if (SiteSetting.ai_bot_allowed_groups_map & post.user.group_ids).blank?
return false if post.custom_fields[BYPASS_AI_REPLY_CUSTOM_FIELD].present?
true
end
def schedule_bot_reply(post)
persona_id =
DiscourseAi::AiBot::Personas::Persona.system_personas[bot.persona.class] ||
bot.persona.class.id
::Jobs.enqueue(
:create_ai_reply,
post_id: post.id,
bot_user_id: bot.bot_user.id,
persona_id: persona_id,
)
end
def context(topic)
{
site_url: Discourse.base_url,
site_title: SiteSetting.title,
site_description: SiteSetting.site_description,
time: Time.zone.now,
participants: topic.allowed_users.map(&:username).join(", "),
}
end
def publish_update(bot_reply_post, payload)
payload = { post_id: bot_reply_post.id, post_number: bot_reply_post.post_number }.merge(
payload,
)
MessageBus.publish(
"discourse-ai/ai-bot/topic/#{bot_reply_post.topic_id}",
payload,
user_ids: bot_reply_post.topic.allowed_user_ids,
max_backlog_size: 2,
max_backlog_age: 60,
)
end
end
end
end