RagLeap Core: 46 built-in AI employee roles, self-hosted and MIT licensed

Dragan Petkovic, an AI solution builder and full-stack developer, recently named RagLeap Core in a German-language LinkedIn post about open-source alternatives to commercial AI assistants. This post explains the parts of that description that concern us: what RagLeap Core is, which 46 roles it ships with, and what you need to run it yourself.

What was said

In his post, Dragan describes RagLeap Core as an open-source platform with 46 role-based AI employees, from CEO to HR to recruiter, that is self-hosted, MIT licensed and needs no license key. Read his post on LinkedIn. The rest of that post is about Microsoft Copilot; we do not vouch for those claims. He also says models can be run locally for free, and we clarify that point below.

What RagLeap Core is

RagLeap Core is an open-source, self-hosted system for AI employees that already know your documents. It is MIT licensed, has no account and no license key, and runs on your own server with Docker Compose. The stack contains Postgres with pgvector, the app, a voice service, Neo4j 5, Redis 7 and a worker. Channel adapters exist for WhatsApp, Telegram, Discord and voice (Twilio).

The 46 built-in roles

The role list in release v0.7.5 has 47 entries: 46 built-in roles plus custom, which lets you define your own. These are the identifiers used in the code, grouped by us for readability.

Core business (9): manager, secretary, ceo, sales, support, hr, finance, marketing, operations
People and community (4): recruiter, volunteer_coordinator, membership_agent, admissions_agent
Sales, property and vehicles (6): real_estate_agent, vehicle_sales_agent, b2b_lead_qualifier, franchise_inquiry_agent, property_management_agent, coworking_space_agent
Hospitality, travel and bookings (6): hospitality_agent, event_planner, travel_agent, photography_booking_agent, salon_spa_agent, fitness_studio_agent
Money, compliance and intake (8): collections_agent, compliance_officer, insurance_agent, tax_preparation_intake, legal_intake, immigration_intake, healthcare_intake, veterinary_intake
Operations and logistics (8): procurement_agent, inventory_agent, logistics_agent, warehouse_operations_agent, delivery_dispatch_agent, returns_refunds_agent, warranty_claims_agent, auto_service_agent
IT, content, data and education (5): it_helpdesk, content_writer, social_media_manager, data_analyst, tutoring_agent

What you need to run it

The quickstart in the README is one command. It checks for Docker, clones the repository, creates a .env file and pauses so you can add your Gemini API key; you then run it again.

curl -fsSL https://raw.githubusercontent.com/antonyrag/ragleap-core/main/install.sh | bash

Or do it by hand: git clone the repository, copy .env.example to .env, add the key and run docker compose up --build -d. The app then answers at http://localhost:8000, with API docs at /docs and a health check at /health. The install guide has the details: docs.ragleap.com/install.

One point worth being precise about: "run models locally for free"

Chat can run on a local model. With the Ollama provider you set only OLLAMA_MODEL and need no key. But embeddings always use Gemini, so a Gemini API key is required whichever chat provider you choose, and the text sent for embedding goes to Google. Your data is stored on your own infrastructure, and text leaves it only for the AI providers you configure: your chat model, and Gemini for embeddings. Chat itself supports 19 providers, including Gemini, Anthropic, OpenAI, Mistral, Groq and any OpenAI-compatible endpoint. See the providers page.

Safety defaults

Autonomy is off by default. The AI acts only when the owner starts it. In approval mode it proposes an action and the owner approves or rejects it. Full autonomy is an explicit opt-in.
Sensitive roles are capped. Roles in sensitive domains such as legal, medical, tax, immigration, insurance and compliance are forced from full autonomy down to approval mode.
The API is local by default. It binds to 127.0.0.1, and it has no login unless you set RAGLEAP_API_KEY. Set a key before exposing it.

The AI employees page in the docs explains roles and autonomy in detail. The source is on GitHub, and our other write-up from the community is about short-query language detection.