discourse-ai/app/models/ai_tool.rb
Sam b5d393b4bc
FIX: custom tools incorrectly setting all fields to blank enum (#1385)
Previous to this change, enum was set to [] which broke all non
enum tools
2025-05-30 17:12:24 +10:00

483 lines
19 KiB
Ruby

# frozen_string_literal: true
class AiTool < ActiveRecord::Base
validates :name, presence: true, length: { maximum: 100 }, uniqueness: true
validates :tool_name, presence: true, length: { maximum: 100 }
validates :description, presence: true, length: { maximum: 1000 }
validates :summary, presence: true, length: { maximum: 255 }
validates :script, presence: true, length: { maximum: 100_000 }
validates :created_by_id, presence: true
belongs_to :created_by, class_name: "User"
belongs_to :rag_llm_model, class_name: "LlmModel"
has_many :rag_document_fragments, dependent: :destroy, as: :target
has_many :upload_references, as: :target, dependent: :destroy
has_many :uploads, through: :upload_references
before_update :regenerate_rag_fragments
ALPHANUMERIC_PATTERN = /\A[a-zA-Z0-9_]+\z/
validates :tool_name,
format: {
with: ALPHANUMERIC_PATTERN,
message: I18n.t("discourse_ai.tools.name.characters"),
}
validate :validate_parameters_enum
def signature
{
name: function_call_name,
description: description,
parameters: parameters.map(&:symbolize_keys),
}
end
# Backwards compatibility: if tool_name is not set (existing custom tools), use name
def function_call_name
tool_name.presence || name
end
def runner(parameters, llm:, bot_user:, context: nil)
DiscourseAi::Personas::ToolRunner.new(
parameters: parameters,
llm: llm,
bot_user: bot_user,
context: context,
tool: self,
)
end
after_commit :bump_persona_cache
def bump_persona_cache
AiPersona.persona_cache.flush!
end
def regenerate_rag_fragments
if rag_chunk_tokens_changed? || rag_chunk_overlap_tokens_changed?
RagDocumentFragment.where(target: self).delete_all
end
end
def validate_parameters_enum
return unless parameters.is_a?(Array)
parameters.each_with_index do |param, index|
next if !param.is_a?(Hash) || !param.key?("enum")
enum_values = param["enum"]
if enum_values.empty?
errors.add(
:parameters,
"Parameter '#{param["name"]}' at index #{index}: enum cannot be empty",
)
next
end
if enum_values.uniq.length != enum_values.length
errors.add(
:parameters,
"Parameter '#{param["name"]}' at index #{index}: enum values must be unique",
)
end
end
end
def self.preamble
<<~JS
/**
* Tool API Quick Reference
*
* Entry Functions
*
* invoke(parameters): Main function. Receives parameters defined in the tool's signature (Object).
* Must return a JSON-serializable value (e.g., string, number, object, array).
* Example:
* function invoke(parameters) { return { result: "Data processed", input: parameters.query }; }
*
* details(): Optional function. Returns a string (can include basic HTML) describing
* the tool's action after invocation, often using data from the invocation.
* This is displayed in the chat interface.
* Example:
* let lastUrl;
* function invoke(parameters) {
* lastUrl = parameters.url;
* // ... perform action ...
* return { success: true, content: "..." };
* }
* function details() {
* return `Browsed: <a href="${lastUrl}">${lastUrl}</a>`;
* }
*
* Provided Objects & Functions
*
* 1. http
* Performs HTTP requests. Max 20 requests per execution.
*
* http.get(url, options?): Performs GET request.
* Parameters:
* url (string): The request URL.
* options (Object, optional):
* headers (Object): Request headers (e.g., { "Authorization": "Bearer key" }).
* Returns: { status: number, body: string }
*
* http.post(url, options?): Performs POST request.
* Parameters:
* url (string): The request URL.
* options (Object, optional):
* headers (Object): Request headers.
* body (string | Object): Request body. If an object, it's stringified as JSON.
* Returns: { status: number, body: string }
*
* http.put(url, options?): Performs PUT request (similar to POST).
* http.patch(url, options?): Performs PATCH request (similar to POST).
* http.delete(url, options?): Performs DELETE request (similar to GET/POST).
*
* 2. llm
* Interacts with the Language Model.
*
* llm.truncate(text, length): Truncates text to a specified token length based on the configured LLM's tokenizer.
* Parameters:
* text (string): Text to truncate.
* length (number): Maximum number of tokens.
* Returns: string (truncated text)
*
* llm.generate(prompt): Generates text using the configured LLM associated with the tool runner.
* Parameters:
* prompt (string | Object): The prompt. Can be a simple string or an object
* like { messages: [{ type: "system", content: "..." }, { type: "user", content: "..." }] }.
* Returns: string (generated text)
*
* 3. index
* Searches attached RAG (Retrieval-Augmented Generation) documents linked to this tool.
*
* index.search(query, options?): Searches indexed document fragments.
* Parameters:
* query (string): The search query used for semantic search.
* options (Object, optional):
* filenames (Array<string>): Filter search to fragments from specific uploaded filenames.
* limit (number): Maximum number of fragments to return (default: 10, max: 200).
* Returns: Array<{ fragment: string, metadata: string | null }> - Ordered by relevance.
*
* 4. upload
* Handles file uploads within Discourse.
*
* upload.create(filename, base_64_content): Uploads a file created by the tool, making it available in Discourse.
* Parameters:
* filename (string): The desired name for the file (basename is used for security).
* base_64_content (string): Base64 encoded content of the file.
* Returns: { id: number, url: string, short_url: string } - Details of the created upload record.
*
* upload.getUrl(shortUrl): Given a short URL, eg upload://12345, returns the full CDN friendly URL of the upload.
* 5. chain
* Controls the execution flow.
*
* chain.setCustomRaw(raw): Sets the final raw content of the bot's post and immediately
* stops the tool execution chain. Useful for tools that directly
* generate the full response content (e.g., image generation tools attaching the image markdown).
* Parameters:
* raw (string): The raw Markdown content for the post.
* Returns: void
*
* 6. discourse
* Interacts with Discourse specific features. Access is generally performed as the SystemUser.
*
* discourse.search(params): Performs a Discourse search.
* Parameters:
* params (Object): Search parameters (e.g., { search_query: "keyword", with_private: true, max_results: 10 }).
* `with_private: true` searches across all posts visible to the SystemUser. `result_style: 'detailed'` is used by default.
* Returns: Object (Discourse search results structure, includes posts, topics, users etc.)
*
* discourse.getPost(post_id): Retrieves details for a specific post.
* Parameters:
* post_id (number): The ID of the post.
* Returns: Object (Post details including `raw`, nested `topic` object with ListableTopicSerializer structure) or null if not found/accessible.
*
* discourse.getTopic(topic_id): Retrieves details for a specific topic.
* Parameters:
* topic_id (number): The ID of the topic.
* Returns: Object (Topic details using ListableTopicSerializer structure) or null if not found/accessible.
*
* discourse.getUser(user_id_or_username): Retrieves details for a specific user.
* Parameters:
* user_id_or_username (number | string): The ID or username of the user.
* Returns: Object (User details using UserSerializer structure) or null if not found.
*
* discourse.getPersona(name): Gets an object representing another AI Persona configured on the site.
* Parameters:
* name (string): The name of the target persona.
* Returns: Object { respondTo: function(params) } or null if persona not found.
* respondTo(params): Instructs the target persona to generate a response within the current context (e.g., replying to the same post or chat message).
* Parameters:
* params (Object, optional): { instructions: string, whisper: boolean }
* Returns: { success: boolean, post_id?: number, post_number?: number, message_id?: number } or { error: string }
*
* discourse.createChatMessage(params): Creates a new message in a Discourse Chat channel.
* Parameters:
* params (Object): { channel_name: string, username: string, message: string }
* `channel_name` can be the channel name or slug.
* `username` specifies the user who should appear as the sender. The user must exist.
* The sending user must have permission to post in the channel.
* Returns: { success: boolean, message_id?: number, message?: string, created_at?: string } or { error: string }
*
* 7. context
* An object containing information about the environment where the tool is being run.
* Available properties depend on the invocation context, but may include:
* post_id (number): ID of the post triggering the tool (if in a Post context).
* topic_id (number): ID of the topic (if in a Post context).
* private_message (boolean): Whether the context is a private message (in Post context).
* message_id (number): ID of the chat message triggering the tool (if in Chat context).
* channel_id (number): ID of the chat channel (if in Chat context).
* user (Object): Details of the user invoking the tool/persona (structure may vary, often null or SystemUser details unless explicitly passed).
* participants (string): Comma-separated list of usernames in a PM (if applicable).
* // ... other potential context-specific properties added by the calling environment.
*
* Constraints
*
* Execution Time: ≤ 2000ms (default timeout in milliseconds) - This timer *pauses* during external HTTP requests or LLM calls initiated via `http.*` or `llm.generate`, but applies to the script's own processing time.
* Memory: 10MB (V8 heap limit)
* Stack Depth: 20 (Marshal stack depth limit for Ruby interop)
* HTTP Requests: 20 per execution
* Exceeding limits will result in errors or termination (e.g., timeout error, out-of-memory error, TooManyRequestsError).
*
* Security
*
* Sandboxed Environment: The script runs in a restricted V8 JavaScript environment (via MiniRacer).
* No direct access to browser or environment, browser globals (like `window` or `document`), or the host system's file system.
* Network requests are proxied through the Discourse backend, not made directly from the sandbox.
*/
JS
end
def self.presets
[
{
preset_id: "browse_web_jina",
name: "Browse Web",
tool_name: "browse_web",
description: "Browse the web as a markdown document",
parameters: [
{ name: "url", type: "string", required: true, description: "The URL to browse" },
],
script: <<~SCRIPT,
#{preamble}
let url;
function invoke(p) {
url = p.url;
result = http.get(`https://r.jina.ai/${url}`);
// truncates to 15000 tokens
return llm.truncate(result.body, 15000);
}
function details() {
return "Read: " + url
}
SCRIPT
},
{
preset_id: "exchange_rate",
name: "Exchange Rate",
tool_name: "exchange_rate",
description: "Get current exchange rates for various currencies",
parameters: [
{
name: "base_currency",
type: "string",
required: true,
description: "The base currency code (e.g., USD, EUR)",
},
{
name: "target_currency",
type: "string",
required: true,
description: "The target currency code (e.g., EUR, JPY)",
},
{ name: "amount", type: "number", description: "Amount to convert eg: 123.45" },
],
script: <<~SCRIPT,
#{preamble}
// note: this script uses the open.er-api.com service, it is only updated
// once every 24 hours, for more up to date rates see: https://www.exchangerate-api.com
function invoke(params) {
const url = `https://open.er-api.com/v6/latest/${params.base_currency}`;
const result = http.get(url);
if (result.status !== 200) {
return { error: "Failed to fetch exchange rates" };
}
const data = JSON.parse(result.body);
const rate = data.rates[params.target_currency];
if (!rate) {
return { error: "Target currency not found" };
}
const rval = {
base_currency: params.base_currency,
target_currency: params.target_currency,
exchange_rate: rate,
last_updated: data.time_last_update_utc
};
if (params.amount) {
rval.original_amount = params.amount;
rval.converted_amount = params.amount * rate;
}
return rval;
}
function details() {
return "<a href='https://www.exchangerate-api.com'>Rates By Exchange Rate API</a>";
}
SCRIPT
summary: "Get current exchange rates between two currencies",
},
{
preset_id: "stock_quote",
name: "Stock Quote (AlphaVantage)",
tool_name: "stock_quote",
description: "Get real-time stock quote information using AlphaVantage API",
parameters: [
{
name: "symbol",
type: "string",
required: true,
description: "The stock symbol (e.g., AAPL, GOOGL)",
},
],
script: <<~SCRIPT,
#{preamble}
function invoke(params) {
const apiKey = 'YOUR_ALPHAVANTAGE_API_KEY'; // Replace with your actual API key
const url = `https://www.alphavantage.co/query?function=GLOBAL_QUOTE&symbol=${params.symbol}&apikey=${apiKey}`;
const result = http.get(url);
if (result.status !== 200) {
return { error: "Failed to fetch stock quote" };
}
const data = JSON.parse(result.body);
if (data['Error Message']) {
return { error: data['Error Message'] };
}
const quote = data['Global Quote'];
if (!quote || Object.keys(quote).length === 0) {
return { error: "No data found for the given symbol" };
}
return {
symbol: quote['01. symbol'],
price: parseFloat(quote['05. price']),
change: parseFloat(quote['09. change']),
change_percent: quote['10. change percent'],
volume: parseInt(quote['06. volume']),
latest_trading_day: quote['07. latest trading day']
};
}
function details() {
return "<a href='https://www.alphavantage.co'>Stock data provided by AlphaVantage</a>";
}
SCRIPT
summary: "Get real-time stock quotes using AlphaVantage API",
},
{
preset_id: "image_generation",
name: "Image Generation (Flux)",
tool_name: "image_generation",
description:
"Generate images using the FLUX model from Black Forest Labs using together.ai",
parameters: [
{
name: "prompt",
type: "string",
required: true,
description: "The text prompt for image generation",
},
{
name: "seed",
type: "number",
required: false,
description: "Optional seed for random number generation",
},
],
script: <<~SCRIPT,
#{preamble}
const apiKey = "YOUR_KEY";
const model = "black-forest-labs/FLUX.1.1-pro";
function invoke(params) {
let seed = parseInt(params.seed);
if (!(seed > 0)) {
seed = Math.floor(Math.random() * 1000000) + 1;
}
const prompt = params.prompt;
const body = {
model: model,
prompt: prompt,
width: 1024,
height: 768,
steps: 10,
n: 1,
seed: seed,
response_format: "b64_json",
};
const result = http.post("https://api.together.xyz/v1/images/generations", {
headers: {
"Authorization": `Bearer ${apiKey}`,
"Content-Type": "application/json",
},
body: JSON.stringify(body),
});
const base64Image = JSON.parse(result.body).data[0].b64_json;
const image = upload.create("generated_image.png", base64Image);
const raw = `\n![${prompt}](${image.short_url})\n`;
chain.setCustomRaw(raw);
return { result: "Image generated successfully", seed: seed };
}
function details() {
return "Generates images based on a text prompt using the FLUX model.";
}
SCRIPT
summary: "Generate image",
},
{ preset_id: "empty_tool", script: <<~SCRIPT },
#{preamble}
function invoke(params) {
// logic here
return params;
}
function details() {
return "Details about this tool";
}
SCRIPT
].map do |preset|
preset[:preset_name] = I18n.t("discourse_ai.tools.presets.#{preset[:preset_id]}.name")
preset
end
end
end
# == Schema Information
#
# Table name: ai_tools
#
# id :bigint not null, primary key
# name :string not null
# description :string not null
# summary :string not null
# parameters :jsonb not null
# script :text not null
# created_by_id :integer not null
# enabled :boolean default(TRUE), not null
# created_at :datetime not null
# updated_at :datetime not null
# rag_chunk_tokens :integer default(374), not null
# rag_chunk_overlap_tokens :integer default(10), not null
# tool_name :string(100) default(""), not null
# rag_llm_model_id :bigint