realtimex_toolkit 1.5.0

Lightweight internal utilities for LLM provider configuration and credential management

RealtimeX Internal Utilities

Lightweight internal library providing utilities for LLM provider configuration and credential management across RealtimeX internal services.

Installation

# Using uv (recommended) for local development
uv pip install -e /path/to/realtimex-toolkit

# Using pip from PyPI
pip install realtimex_toolkit

Quick Start

Agent Flow Management

Work with agent flow data. get_flow_variable now supports dotted paths for nested structures, safe defaults, and fetching the full payload.

import json

from realtimex_toolkit import (
    get_agent_id,
    get_flow_variable,
    get_workspace_data_dir,
    get_workspace_slug,
)

# Access nested variables using dotted notation
user_email = get_flow_variable("user.email")
print(user_email)  # "[e-mail]"

# Provide defaults for missing keys
theme = get_flow_variable("user.theme", "light")
print(theme)  # "light"

# Get the entire payload when you need to inspect everything
variables = get_flow_variable()
print(json.dumps(variables))  # {"time":"12:40:28 PM", ...}

# Get workspace slug (falls back to default when not running inside a flow)
workspace_slug = get_workspace_slug(default_value="workspace-test")
print(workspace_slug)  # "workspace-test" when default is used; otherwise current workspace slug

# Get current agent id (falls back to default when not present)
agent_id = get_agent_id(default_value="agent-test")
print(agent_id)  # "agent-test" when default is used; otherwise current agent id

# Get workspace data directory (creates it if needed)
workspace_data_dir = get_workspace_data_dir(default_workspace_slug=workspace_slug)
print(workspace_data_dir)  # e.g., ~/.realtimex.ai/Resources/server/storage/working-data/<slug>

Agent Memory & Skills (Workspace)

Save workspace-scoped Deep Agent memory and skills under ~/.realtimex.ai/Resources/agent-skills/workspaces/{workspace_slug}/{agent_id}/.

from realtimex_toolkit import save_agent_memory, save_agent_skill

# Overwrite workspace agent.md
path = save_agent_memory("workspace-slug", "agent-id", "# New memory")
# path -> ~/.realtimex.ai/Resources/agent-skills/workspaces/workspace-slug/agent-id/agent.md

# Append to agent.md
save_agent_memory("workspace-slug", "agent-id", "- Added preference", mode="append")

# Overwrite SKILL.md for a skill
skill_path = save_agent_skill(
    "workspace-slug",
    "agent-id",
    "web-research",
    "---\nname: web-research\ndescription: Web research workflow\n---\nSteps...",
)
# skill_path -> ~/.realtimex.ai/Resources/agent-skills/workspaces/workspace-slug/agent-id/skills/web-research/SKILL.md

# Append to SKILL.md
save_agent_skill("workspace-slug", "agent-id", "web-research", "More notes", mode="append")

Gmail MCP Email Lookup

Retrieve the configured Gmail email address for your MCP server using MCP proxy credentials.

from realtimex_toolkit import get_gmail_email

email = get_gmail_email()
print(email)  # e.g., [e-mail]

Credential Management

Retrieve encrypted credentials from the RealtimeX app backend and decrypt them for use across RealtimeX ecosystem services.

from realtimex_toolkit import get_credential

# Simplest usage - just the credential ID
# Connects to local RealtimeX backend (http://localhost:3001) by default
credential = get_credential("credential-id")
print(credential["payload"])  # {"name": "API_KEY", "value": "secret-value"}

# With API key for authenticated requests
credential = get_credential("credential-id", api_key="service-api-key")

# With custom backend URL (for non-default configurations)
credential = get_credential(
    "credential-id",
    api_key="service-api-key",
    base_url="http://custom-host:3001"  # Override default localhost:3001
)

# For long-running services, use CredentialManager directly
from realtimex_toolkit import CredentialManager

# Connects to http://localhost:3001 by default
manager = CredentialManager(api_key="service-api-key")
try:
    bundle = manager.get("credential-id")
    # Credentials are cached automatically
    bundle_again = manager.get("credential-id")  # Returns cached

    # Force refresh from backend
    fresh_bundle = manager.get("credential-id", force_refresh=True)
finally:
    manager.close()

Configuration:

• base_url: Base URL of the RealtimeX app backend (default: http://localhost:3001)
• api_key: Authentication token for backend API requests (optional)
• Credentials are encrypted using AES-256-CBC and decrypted using keys from ~/.realtimex.ai/Resources/server/.env.development

Return shape (get_credential):

{
    "credential_id": str,
    "name": str,
    "credential_type": str,
    "payload": dict[str, str],
    "metadata": dict | None,
    "updated_at": str | None,
}

Supported LLM Providers

• Major Providers: OpenAI, Anthropic, Azure OpenAI
• Cloud AI: Google Gemini, AWS Bedrock
• Alternative APIs: Groq, Cohere, Mistral, Perplexity
• Open Source Aggregators: Open Router, Together AI, Fireworks AI
• Emerging: DeepSeek, xAI, Novita
• Local Deployment: Ollama, LocalAI, LM Studio, KoboldCPP
• Custom: Generic OpenAI, LiteLLM, Nvidia NIM, Hugging Face

Development

# Install with dev dependencies
uv pip install -e ".[dev]"

# Run tests
pytest

# Run type checking
mypy src/realtimex_toolkit

# Format & lint
ruff check src/realtimex_toolkit tests
ruff format src/realtimex_toolkit tests

Architecture

• realtimex_toolkit.llm: LLM provider configuration utilities
• realtimex_toolkit.credentials: Secure credential retrieval and decryption
• realtimex_toolkit.api: HTTP client with retry logic and error mapping
• realtimex_toolkit.utils: Internal utilities (path resolution, logging)

License

Proprietary - Internal use only