PhoenixAI Store

Persistence, memory management, guardrails, cost tracking, and an audit event log for PhoenixAI conversations.

Features

• Conversation Persistence — ETS (in-memory) and Ecto (PostgreSQL/SQLite) adapters
• Memory Strategies — Sliding window, token-aware truncation, pinned messages
• Guardrails — Token budgets, cost budgets, and rate limiting before AI calls
• Cost Tracking — Per-conversation and per-user cost accumulation with Decimal precision
• Event Log — Append-only audit trail with cursor pagination and PII redaction
• Single-Function Pipeline — converse/3 handles load → memory → guardrails → AI → save → track

Installation

Add phoenix_ai_store to your list of dependencies in mix.exs:

def deps do
  [
    {:phoenix_ai_store, "~> 0.1.0"}
  ]
end

Quick Start

# Add to your supervision tree
children = [
  {PhoenixAI.Store, name: :my_store, adapter: PhoenixAI.Store.Adapters.ETS}
]

# Create a conversation
{:ok, conv} = PhoenixAI.Store.save_conversation(
  %PhoenixAI.Store.Conversation{title: "My Chat"},
  store: :my_store
)

# Run the full pipeline
{:ok, response} = PhoenixAI.Store.converse(
  conv.id,
  "Hello!",
  provider: :openai,
  model: "gpt-4o",
  api_key: System.get_env("OPENAI_API_KEY"),
  store: :my_store
)

Documentation

• Getting Started — Installation and first conversation
• Adapters Guide — ETS vs Ecto, custom adapters
• Memory & Guardrails — Context window management
• Telemetry & Events — Automatic event capture

License

MIT — see LICENSE for details.