EPI

Capture any AI agent run into one portable .epi file you can open, share, and verify anywhere.

The forensic bug report artifact for AI systems. No cloud. No login. No internet required.

Reference implementation of EPI (Evidence Packaged Infrastructure) v4.0.3 — the open format for packaging AI execution as tamper-evident, portable evidence.

Install

pip install epi-recorder

Get Started in 60 Seconds

epi demo

Runs a sample refund workflow and gives you the full developer repro loop:

1. Capture an AI agent run into a portable .epi artifact.
2. View the case in a browser with Overview, Evidence, Policy, and Trust.
3. Verify the cryptographic integrity of the file.

Core API

from epi_recorder import record, wrap_openai
from openai import OpenAI

client = wrap_openai(OpenAI())

with record("my_agent.epi", workflow_name="Investigation"):
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": "Analyze this data."}]
    )

Open and verify the artifact:

epi view my_agent.epi    # Browser review (offline)
epi verify my_agent.epi  # Ed25519 integrity check

The .epi Format

my_agent.epi
|- EPI1 header            # outer identity, payload length, SHA-256
`- ZIP evidence payload
   |- manifest.json        # metadata + Ed25519 signature + content hashes
   |- steps.jsonl          # execution timeline (NDJSON)
   |- environment.json     # runtime snapshot
   `- viewer.html          # self-contained offline viewer

flowchart LR
    A["Agent Code"] -->|"record()"| B["Capture Layer"]
    B --> C["SQLite WAL"]
    C --> D["ZIP Payload Builder"]
    E["Private Key"] -->|"Ed25519 Sign Manifest"| D
    D -->|"Wrap with EPI1 Envelope"| G["agent.epi"]

See EPI Specification for details. EPI supports evidence workflows for EU AI Act, FDA, and financial services; it does not provide legal advice and is not a compliance guarantee.

Framework Integrations

EPI provides native hooks for major AI frameworks:

• LiteLLM: litellm.callbacks = [EPICallback()]
• LangChain: ChatOpenAI(..., callbacks=[EPICallbackHandler()])
• Guardrails AI: instrument(output_path="audit.epi")
• OpenTelemetry: setup_epi_tracing(service_name="my-agent")
• pytest: pytest --epi (records signed evidence for failing tests)

See Framework Integrations Guide for details.

CLI Reference

Full reference: docs/CLI.md

Technical Documentation

• EPI Specification: Technical details of the .epi format.
• Policy Guide: How policy, fault analysis, and rulebooks work.
• EU AI Act Evidence Prep: Evidence workflow guide.
• AGT Import Quickstart: AGT -> EPI path.

Contributing

git clone https://github.com/mohdibrahimaiml/epi-recorder.git
cd epi-recorder
pip install -e ".[dev]"
pytest

MIT License. Built by EPI Labs. Making AI agent execution verifiable.