The EPI Lifecycle

Transform any AI Evidence/Workflow into a self-contained, verifiable evidence package.

PKG

0. Install

Get started in seconds. Install the epi-recorder package from PyPI with a single command.

$ pip install epi-recorder

New in v2.8.7: Native OS File Associations & Framework integrations - LiteLLM (100+ providers), LangChain, pytest plugin, GitHub Action, OpenTelemetry, and streaming support. See Get Started for details.

REC

1. Record Shipped

Explicit "Flight Recorder" for your AI. Use log_llm_call() or wrappers to capture inputs, calls, and outputs into a single sealed file.

$ epi record script.py
VERIFY

2. Verify

The EPI Verifier replays the file in a sandbox. It proves the output was generated by the exact code and inputs claimed.

$ epi verify scan_01.epi
SHARE

3. Share Evidence

Send the .epi file to auditors, investors, or regulators. It's a portable proof of work that doesn't need your secrets.

(Drag & Drop File Here)
VIEW

4. View Shipped

Self-contained HTML viewer. Open .epi files like PDFs. No execution, no risk. Human-readable timeline of events.

$ epi view output.epi

Under the Hood

Enterprise-grade architecture. No proprietary lock-in.

ZIP-Based Container

Self-contained .epi files. Store evidence + viewer in one portable archive.

Ed25519 Signatures

Same crypto as Signal & SSH. Tamper-proof, military-grade security.

Content-Addressed Storage

SHA-256 hashing ensures file integrity. Deduplication by design.

Multi-Provider Support

OpenAI, Anthropic, Ollama, Google Gemini. Automatic capture.

Offline Replay

Cached LLM responses enable deterministic re-runs without API calls.

Embedded Viewer

Static HTML/JS timeline. No server required. Works like a PDF.

The Trust Architecture

How we turn code into verifiable evidence.

Input

Code Execution
->

Integrity

SHA-256 Hash
->

Identity

Ed25519 Sign
->

Result

Verified .epi