runfile-ai 0.3.0

Runfile SDK — tamper-evident audit capture for AI agents (LangGraph, OpenAI Agents, Claude Agent SDK, MCP).

runfile-ai

The Runfile SDK for Python. Tamper-evident audit capture for AI agents.

pip install runfile-ai
# framework extras:
pip install 'runfile-ai[langgraph]'   # or [openai-agents], [anthropic], [mcp]

import os
import runfile_ai
from runfile_ai.integrations import langgraph as runfile_langgraph

runfile_ai.init(api_key=os.environ["RUNFILE_API_KEY"], environment="production")

graph = runfile_langgraph.instrument(
    graph, agent_identity="did:web:bank.com:agents:loan-triage:v2"
)

Configuration

Explicit args take precedence over environment variables, which take precedence over defaults:

Arg	Env var	Default
api_key	RUNFILE_API_KEY	— (required)
region	RUNFILE_REGION	eu-west-2
environment	RUNFILE_ENVIRONMENT	production
disabled	RUNFILE_DISABLED	false

base_url is derived as https://api.<region>.runfile.ai unless overridden. RUNFILE_DISABLED=1 makes the SDK a silent no-op (handy for local dev). So runfile_ai.init() with RUNFILE_API_KEY set is enough.

• Import name: runfile_ai (distribution: runfile-ai).
• Wire sdk.name: runfile-ai.
• Python: 3.11+.
• Schema dependency: runfile-ai-schemas (PyPI), pinned >=0.5,<1.0.

🚧 Scaffold — module skeletons are in place; logic is being filled in. See the repo root README.md and docs/STATUS.md.

Local development

pip install -e '.[dev]'
ruff check .
mypy runfile_ai
pytest