omnismart-personas

Role-based, data-scoped LLM persona templates + router for production analytics copilots. The pattern behind IntelAI's 9-persona RAG copilot, extracted as a tiny, zero-dependency, pure-Python library.

A persona pairs a system prompt with a data-access scope (which business domains it may see) and a sampling temperature — so the same retrieval system can answer through different role-conditioned prompts and data filters ("persona-routed RAG").

Install

pip install omnismart-personas

Use

from omnismart_personas import persona_for_role, scope_records, build_rag_prompt

persona = persona_for_role("cfo")          # → CFO persona (scope: Finance, Growth)
persona.can_access("People")               # False — out of scope

rows = [
    {"category": "Finance", "metric": "gross_margin", "value": 0.42},
    {"category": "People",  "metric": "headcount",    "value": 220},
]
visible = scope_records(persona, rows)     # drops the People row

prompt = build_rag_prompt(
    persona,
    "What drove the Q1 gross-margin change?",
    [f"{r['metric']}={r['value']}" for r in visible],
    language="en",
)
# → send `prompt` to any LLM

API

• list_personas() → the 9 persona keys
• get_persona(name) / persona_for_role(role) → a Persona
• Persona(name, display_name, system_prompt, allowed_tools, data_access, temperature) — immutable; .can_access(domain)
• scope_records(persona, records, domain_key="category") → records in scope
• build_system_prompt(persona, *, language="en", extra=None)
• build_rag_prompt(persona, query, snippets, *, language="en")

Personas: ceo, cfo, cto, coo, chro, esg, risk, analyst, general.

LangChain

Use the personas in any LangChain RAG chain — pip install "omnismart-personas[langchain]":

from omnismart_personas import persona_for_role
from omnismart_personas.langchain import persona_chat_prompt, persona_retriever_filter
from langchain_openai import ChatOpenAI
from langchain_core.output_parsers import StrOutputParser

persona = persona_for_role("cfo")
prompt = persona_chat_prompt(persona)               # system = scope; {context} + {question}
chain = prompt | ChatOpenAI(model="gpt-4o-mini") | StrOutputParser()

docs = persona_retriever_filter(persona, retriever.invoke(q))   # RBAC on retrieved docs
answer = chain.invoke({"context": "\n".join(d.page_content for d in docs), "question": q})

persona_retriever_filter drops Documents whose metadata["category"] is outside the persona's data_access scope — the same role boundaries, enforced in your LangChain pipeline.

Test

pip install -e ".[test]" && pytest

MIT licensed. Part of the IntelAI project.