Ejentum Python SDK

The official Python SDK for Ejentum — Deterministic Reasoning-Augmented Retrieval (RAR) Cognitive Infrastructure.

Ejentum indexes 300+ deterministic mental models and cognitive scaffolds. In <80ms, it retrieves and injects the optimal reasoning scaffold into your LLM prompt, eliminating hallucinated leaps, sycophantic shortcuts, and premature convergence in production AI agents.

⚡ Installation

pip install ejentum

Zero dependencies required. Uses Python standard library only.

🔑 Authentication

Export your API key in your environment (or pass api_key="..." directly to the client):

export EJENTUM_API_KEY="ej_live_your_key_here"

Need an API key? Get instant access at ejentum.com/pricing.

🚀 Quickstart

1. Basic Scaffold Retrieval

from ejentum import Ejentum

client = Ejentum()

# Retrieve the deterministic cognitive scaffold for any query (<80ms)
scaffold = client.scaffold("diagnose PostgreSQL high CPU and lock contention")
print(scaffold)

2. Auto-Augment a Prompt

# Automatically prepends the scaffold instruction block to your task prompt
prompt = "Design a failover strategy for our payment processing service."
augmented_prompt = client.augment(prompt, mode="reasoning")

print(augmented_prompt)
# Output:
# Apply the cognitive scaffold below, then answer the user's task.
# [COGNITIVE SCAFFOLD]
# ... [Deterministic Causal / Resiliency Scaffold] ...
# [END SCAFFOLD]
# Design a failover strategy for our payment processing service.

🔌 1-Line Drop-in Wrappers

OpenAI Drop-in

Wrap your existing OpenAI() client. All chat.completions.create calls will automatically fetch and inject the optimal Ejentum scaffold before invoking OpenAI:

from openai import OpenAI
from ejentum import wrap_openai

# Wrap your existing client in 1 line:
client = wrap_openai(OpenAI())

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "user", "content": "Diagnose silent failure in checkout webhooks"}
    ],
    ejentum_mode="reasoning"  # Optional: defaults to "reasoning"
)

print(response.choices[0].message.content)

Anthropic Drop-in

from anthropic import Anthropic
from ejentum import wrap_anthropic

client = wrap_anthropic(Anthropic())

response = client.messages.create(
    model="claude-3-7-sonnet-20250219",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Refactor auth token refresh to prevent race condition"}
    ]
)

print(response.content[0].text)

LangChain / Agent Loop Integration

from ejentum import Ejentum
from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI

ejentum = Ejentum()

def scaffold_runnable(inputs):
    scaffold = ejentum.scaffold(inputs["task"])
    return {
        "system_scaffold": f"Apply this scaffold:\n{scaffold}",
        "task": inputs["task"]
    }

prompt = ChatPromptTemplate.from_messages([
    ("system", "{system_scaffold}"),
    ("user", "{task}")
])

chain = scaffold_runnable | prompt | ChatOpenAI(model="gpt-4o")
result = chain.invoke({"task": "Plan a zero-downtime database migration for 50M rows."})

🛠️ API Reference

Ejentum(api_key=None, base_url="https://api.ejentum.com/logicv1/", timeout=10.0)

• scaffold(query: str, mode="reasoning") -> str: Returns the raw cognitive scaffold string.
• augment(prompt: str, mode="reasoning") -> str: Returns the prompt augmented with cognitive scaffolding.
• augment_messages(messages: list[dict], mode="reasoning") -> list[dict]: Augments an OpenAI/Anthropic messages list in-place or prepends to the system message.
• get_ability(query: str, mode="reasoning") -> dict: Returns the full metadata payload including ability_id and matched domain.

📄 License

MIT © Ejentum.com