agenko

The deterministic trust boundary for AI agents.

One line gives every agent action a cryptographic receipt and a deterministic gate before any irreversible move. No LLM inference — measured facts, provable audit, OTEL-native.

Installation

pip install agenko

Quick Start

import agenko

agenko.init()  # instrument is a backward-compat alias of init

@agenko.check
def handle_ticket(ticket):
    result = process(ticket)
    return result  # result._agenko_receipt holds claimed, actual, verdict, hash

@agenko.approve(when="irreversible")
def delete_account(user_id):
    db.delete_user(user_id)

Set AGENKO_API_KEY (and optionally AGENKO_HOST) in your environment — init() reads them and never throws.

What Agenko Provides

Why deterministic matters

Most tools detect silent failures via LLM scoring or semantic drift — useful but not provable. Agenko's receipt is measured facts. We never sell an inference as a measurement. Grounding scores are always estimated: true.

Framework support

init() auto-detects and instruments loaded SDKs — OpenAI, Anthropic, LangChain, Mistral, Cohere, Bedrock, LiteLLM, LlamaIndex, CrewAI, Google GenAI. Import your framework, then call init():

import openai
import agenko

agenko.init()  # patches openai if already imported

Optional drop-in modules: agenko.openai, agenko.anthropic, agenko.langchain, and others.

Self-Host

git clone https://github.com/agenko-ai/agenko
cd agenko
cp .env.example .env
make dev

Open http://localhost:3000 for the console.

Links

• GitHub
• Documentation
• PyPI

License

Apache 2.0