FEATURE: search command now support searching in context of user (#610)

This optional feature allows search to be performed in the context
of the user that executed it.

By default we do not allow this behavior cause it means llm gets
access to potentially secure data.
This commit is contained in:
Sam 2024-05-10 11:32:34 +10:00 committed by GitHub
parent 514823daca
commit 61890b667c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
6 changed files with 110 additions and 32 deletions

View File

@ -1,17 +1,44 @@
import Component from "@glimmer/component";
import { Input } from "@ember/component"; import { Input } from "@ember/component";
import { on } from "@ember/modifier";
import { action } from "@ember/object";
const AiPersonaCommandOptionEditor = <template> export default class AiPersonaCommandOptionEditor extends Component {
<div class="control-group ai-persona-command-option-editor"> get isBoolean() {
<label> return this.args.option.type === "boolean";
{{@option.name}} }
</label>
<div class="">
<Input @value={{@option.value.value}} />
</div>
<div class="ai-persona-command-option-editor__instructions">
{{@option.description}}
</div>
</div>
</template>;
export default AiPersonaCommandOptionEditor; get selectedValue() {
return this.args.option.value.value === "true";
}
@action
onCheckboxChange(event) {
this.args.option.value.value = event.target.checked ? "true" : "false";
}
<template>
<div class="control-group ai-persona-command-option-editor">
<label>
{{@option.name}}
</label>
<div class="">
{{#if this.isBoolean}}
<input
type="checkbox"
checked={{this.selectedValue}}
{{on "click" this.onCheckboxChange}}
/>
{{@option.description}}
{{else}}
<Input @value={{@option.value.value}} />
{{/if}}
</div>
{{#unless this.isBoolean}}
<div class="ai-persona-command-option-editor__instructions">
{{@option.description}}
</div>
{{/unless}}
</div>
</template>
}

View File

@ -205,6 +205,9 @@ en:
searching: "Searching for: '%{query}'" searching: "Searching for: '%{query}'"
command_options: command_options:
search: search:
search_private:
name: "Search Private"
description: "Include all topics user has access to in search results (by default only public topics are included)"
max_results: max_results:
name: "Maximum number of results" name: "Maximum number of results"
description: "Maximum number of results to include in the search - if empty default rules will be used and count will be scaled depending on model used. Highest value is 100." description: "Maximum number of results to include in the search - if empty default rules will be used and count will be scaled depending on model used. Highest value is 100."

View File

@ -83,7 +83,11 @@ module DiscourseAi
end end
def accepted_options def accepted_options
[option(:base_query, type: :string), option(:max_results, type: :integer)] [
option(:base_query, type: :string),
option(:max_results, type: :integer),
option(:search_private, type: :boolean),
]
end end
end end
@ -106,12 +110,18 @@ module DiscourseAi
search_string = "#{search_string} #{options[:base_query]}" search_string = "#{search_string} #{options[:base_query]}"
end end
safe_search_string = search_string.to_s
guardian = nil
if options[:search_private] && context[:user]
guardian = Guardian.new(context[:user])
else
guardian = Guardian.new
safe_search_string += " status:public"
end
results = results =
::Search.execute( ::Search.execute(safe_search_string, search_type: :full_page, guardian: guardian)
search_string.to_s + " status:public",
search_type: :full_page,
guardian: Guardian.new(),
)
# let's be frugal with tokens, 50 results is too much and stuff gets cut off # let's be frugal with tokens, 50 results is too much and stuff gets cut off
max_results = calculate_max_results(llm) max_results = calculate_max_results(llm)
@ -129,10 +139,10 @@ module DiscourseAi
posts = posts[0..results_limit.to_i - 1] posts = posts[0..results_limit.to_i - 1]
if should_try_semantic_search if should_try_semantic_search
semantic_search = DiscourseAi::Embeddings::SemanticSearch.new(Guardian.new()) semantic_search = DiscourseAi::Embeddings::SemanticSearch.new(guardian)
topic_ids = Set.new(posts.map(&:topic_id)) topic_ids = Set.new(posts.map(&:topic_id))
search = ::Search.new(search_string, guardian: Guardian.new) search = ::Search.new(search_string, guardian: guardian)
results = nil results = nil
begin begin

View File

@ -66,14 +66,20 @@ module DiscourseAi
end end
def options def options
self result = HashWithIndifferentAccess.new
.class self.class.accepted_options.each do |option|
.accepted_options val = @persona_options[option.name]
.reduce(HashWithIndifferentAccess.new) do |memo, option| if val
val = @persona_options[option.name] case option.type
memo[option.name] = val if val when :boolean
memo val = val == "true"
when :integer
val = val.to_i
end
result[option.name] = val
end end
end
result
end end
def chain_next_response? def chain_next_response?

View File

@ -29,13 +29,29 @@ RSpec.describe DiscourseAi::AiBot::Tools::Search do
Fabricate(:topic, category: category, tags: [tag_funny, tag_sad, tag_hidden]) Fabricate(:topic, category: category, tags: [tag_funny, tag_sad, tag_hidden])
end end
fab!(:user)
fab!(:group)
fab!(:private_category) do
c = Fabricate(:category_with_definition)
c.set_permissions(group => :readonly)
c.save
c
end
before { SiteSetting.ai_bot_enabled = true } before { SiteSetting.ai_bot_enabled = true }
describe "#invoke" do describe "#invoke" do
it "can retrieve options from persona correctly" do it "can retrieve options from persona correctly" do
persona_options = { "base_query" => "#funny" } persona_options = {
"base_query" => "#funny",
"search_private" => "true",
"max_results" => "10",
}
search_post = Fabricate(:post, topic: topic_with_tags) search_post = Fabricate(:post, topic: topic_with_tags)
private_search_post = Fabricate(:post, topic: Fabricate(:topic, category: private_category))
private_search_post.topic.tags = [tag_funny]
private_search_post.topic.save!
_bot_post = Fabricate(:post) _bot_post = Fabricate(:post)
@ -46,18 +62,28 @@ RSpec.describe DiscourseAi::AiBot::Tools::Search do
bot_user: bot_user, bot_user: bot_user,
llm: llm, llm: llm,
context: { context: {
user: user,
}, },
) )
expect(search.options[:base_query]).to eq("#funny")
expect(search.options[:search_private]).to eq(true)
expect(search.options[:max_results]).to eq(10)
results = search.invoke(&progress_blk) results = search.invoke(&progress_blk)
expect(results[:rows].length).to eq(1) expect(results[:rows].length).to eq(1)
GroupUser.create!(group: group, user: user)
results = search.invoke(&progress_blk)
expect(results[:rows].length).to eq(2)
search_post.topic.tags = [] search_post.topic.tags = []
search_post.topic.save! search_post.topic.save!
# no longer has the tag funny # no longer has the tag funny, but secure one does
results = search.invoke(&progress_blk) results = search.invoke(&progress_blk)
expect(results[:rows].length).to eq(0) expect(results[:rows].length).to eq(1)
end end
it "can handle no results" do it "can handle no results" do

View File

@ -75,6 +75,12 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
"description" => "description" =>
I18n.t("discourse_ai.ai_bot.command_options.search.max_results.description"), I18n.t("discourse_ai.ai_bot.command_options.search.max_results.description"),
}, },
"search_private" => {
"type" => "boolean",
"name" => I18n.t("discourse_ai.ai_bot.command_options.search.search_private.name"),
"description" =>
I18n.t("discourse_ai.ai_bot.command_options.search.search_private.description"),
},
}, },
) )