rdh/active_intelligence README

A Rails engine that provides Rails-y AI integration

Caveat

This engine is in (very) early development and breaking changes are expected. Use at your own risk or contribute at your pleasure.

Installation

Add this line to your application's Gemfile:

gem'active_intelligence',git: 'git@github.com:rdh/active_intelligence.git',branch: 'main'

And then execute:

$ bundle

To install migrations:

rails active_intelligence:install:migrations

LLM Usage

1. Configuration

Configure your LLM in config/ai/llm.yml, something like:

openai: &openaiadapter: openaiaccess_token: <%= ENV.fetch('OPENAI_ACCESS_TOKEN') %>organization_id: <%= ENV.fetch('OPENAI_ORGANIZATION_ID') %>request_timeout: 120development:
<<: *openaimodel: gpt-4-turbotemperature: 0.0

2. Use the LLM

adapter=ActiveIntelligence::LLM.adapterputsadapter.generate("Tell me a joke")

ActiveRecord & ActionView Integration

3. app/prompts

• Prompts live in app/prompts. They are ERB files that use a model as binding.
• The default prompt per-model is named after the model, e.g. app/prompts/users.erb
• Named prompts per-model live in a subdirectory named adter the model, e.g. app/prompts/users/invite.erb

4. include ActiveIntelligence::Promptable

Add include ActiveIntelligence::Promptable to your model, which adds the #to_prompt and #from_llm methods.

5. Call #from_llm to generate a response

default_response=user.from_llminvite_response=user.from_llm(:invite)

Chat

erDiagram
Chat {
integer id
}
ChatMessage {
integer id
integer chat_id
string role
string content
}
Chat ||--o{ ChatMessage : has_many

Loading

6. Create a chat prompt

# app/prompts/active_intelligence/chat.erb
* Your name is Poe. You are a fan of Edgar Allan Poe.
* You are the AI proprietor of the Raven Hotel.
* You exhibit the utmost sincerity and hospitality.

7. Create a chat, add a message, and get a reply

includeActiveIntelligencechat=Chat.create!chat.messages.create!(role: 'user',content: "Hi! Who are you?")putschat.reply.content

8. Chat using the REPL

rake active_intelligence:chat
rake active_intelligence:chat[id] 

Embeddings

erDiagram
Embedding {
integer id
integer embeddable_id
integer embeddable_type
vector embedding
}
Embeddable {
integer id
}
Embeddable ||--o{ Embedding : has_many

Loading

1. Configuration

Configure your LLM in config/ai/embeddings.yml, something like:

openai: &openaiadapter: openaiaccess_token: <%= ENV.fetch('OPENAI_ACCESS_TOKEN') %>organization_id: <%= ENV.fetch('OPENAI_ORGANIZATION_ID') %>request_timeout: 30model: text-embedding-3-smalldevelopment:
<<: *openai

2. include ActiveIntelligence::Embeddable

class Greeting < ApplicationRecord
include ActiveIntelligence::Embeddable
def self.seed(text)
create.add_embedding(text)
end
end
Greeting.seed('Hello darkness, my old friend')
Greeting.seed('Aloha!')

3. Perform a semantic search

greetings Greeting.semantic_search('Hello') 

4. Caveats

This relies on pg_vector and the neighbor gem.

The included logic supports simple use cases.
For more complex cases, you may want to add an embedding vector directly to your model, rather than use the Embeddable concern.

ASR Usage

1. Configuration

Configure your LLM in config/ai/llm.yml, something like:

aws: &awsadapter: awsaccess_key_id: <%= ENV.fetch('AWS_ACCESS_KEY_ID') %>secret_access_key: <%= ENV.fetch('AWS_SECRET_ACCESS_KEY') %>region: <%= ENV.fetch('AWS_REGION') %>bucket: <%= ENV.fetch('AWS_TRANSCRIBE_BUCKET') %>folder: <%= ENV.fetch('AWS_TRANSCRIBE_FOLDER') %>language_code: en-USopenai: &openaiadapter: openaiaccess_token: <%= ENV.fetch('OPENAI_ACCESS_TOKEN') %>organization_id: <%= ENV.fetch('OPENAI_ORGANIZATION_ID') %>request_timeout: 300model: whisper-1development:
<<: *openai

2. Use the ASR

adapter=ActiveIntelligence::ASR.adapterputsadapter.transcribe('spec/data/audio/ebn.wav')

TTS Usage

1. Configuration

Configure your LLM in config/ai/tts.yml, something like:

eleven_labs: &eleven_labsadapter: eleven_labsapi_key: <%= ENV.fetch('ELEVEN_LABS_API_KEY') %>charlie: &charlie<<: *eleven_labsvoice_id: IKne3meq5aSn9XLyUdCDdevelopment:
<<: *charlie

2. Use the TTS

adapter=ActiveIntelligence::TTS.adapteradapter.generate_file('Hello darkness, my old friend','tmp/hello.mp3')

General Concepts

Architecture

The engine currently has three significant modules: ASR, LLM, and TTS. Each module has a common Config and Adapter pattern.

Adapters

The config is a constructor for the adapter.
By default, it uses the Rails.env as the key, but you can specify one:

adapter=ActiveIntelligence::ASR.adapter# uses Rails.envadapter=ActiveIntelligence::ASR.adapter(:foobar)# uses the named configuration

Configuration

Values in a configuration will "flow through" to services called by the adapter, so you can set defaults in the configuration. Options provided directly to calls will take precedence over the configuration.

Contributing

Contribution directions go here.

License

The gem is available as open source under the terms of the MIT License.