RASA – Role-Aligned Software Architecture

Persona-Driven, Memory-Integrated AI Framework for Next-Gen AI Agents

---

🚀 What is RASA?

RASA is a modular Python framework for building advanced, persona-driven, memory-aware AI agents.
It empowers teams to create agents with unique "personality" and reasoning flows, multi-layered memory, and domain-specific intelligence—operationalized as APIs, command-line tools, or direct Python modules.

• Persona-based: Rich agent personas defined in YAML—each with frames (cognitive layers), operators (reasoning steps), and metadata.
• Memory-integrated: First-class support for stateless, short-term, session, and long-term memory (e.g. Redis, vector DB).
• Extensible: Plug in new personas, frames, operators, tools, or LLMs with minimal code.
• Multi-interface: Access RASA via FastAPI server, CLI, or import as a Python library.

---

📁 Project Structure

rasa/
  core/           # Core logic: runners, memory, agent base classes
  api/            # FastAPI app (main entrypoint)
  frames/         # Built-in cognitive frame classes
  operators/      # Built-in operator modules/tools
  llm/            # LLM adapters and config [see ./rasa/llm/LLM_CONFIG.md]
clients/
  rasa.py         # Unified CLI (local+API mode)
  README.md       # CLI usage and tips
apps/
  <persona>/      # Each persona in its own folder
    persona.yaml
    frames/
    operators/
  README.md       # Persona/app usage guide
  PERSONA.md      # Persona YAML schema/standards
tests/
  ...             # Pytest suite for API, CLI, core, personas
README.md         # (You are here - project onboarding!)
ARCHITECTURE.md   # System architecture and extensibility
CHANGELOG.md      # Version history and changes

---

⚡ Quick Start for Users & Developers

1. Clone the repo

git clone https://github.com/vedanta/rasa
cd rasa

2. Set up a Conda Environment (Recommended for All Platforms)

conda create -n rasa python=3.10
conda activate rasa
pip install -r requirements.txt
# (Optional, recommended for devs)
pip install -e .

3. Run the API Server

uvicorn rasa.api.main:app --reload

• Runs at http://localhost:8000 by default.
• Explore docs at http://localhost:8000/docs

4. Try the CLI

python -m clients.rasa list
python -m clients.rasa run --persona travel_concierge --input "Suggest a scenic European trip"

• See full CLI help: python -m clients.rasa --help
• To run against the API:
  python -m clients.rasa --mode api run --persona travel_concierge --input "Ideas for a weekend in Japan"

---

👤 Adding Personas

1. Create a new folder:

   apps/my_persona/
   

2. Write a persona.yaml: See apps/PERSONA.md for the full schema/spec.
3. (Optional) Add custom frames/operators under frames/ and operators/ for persona/domain logic.
4. Test via CLI or API:
   • python -m clients.rasa run --persona my_persona --input "Test prompt"

---

🧪 Testing

• API:
  pytest tests/test_api_master.py -s
• CLI:
  pytest tests/test_cli_end_to_end.py -s
• All:
  pytest

All major endpoints and flows are covered with rich debug output.

---

🛠️ Advanced Usage

• Streaming responses:
  Use --stream in CLI or /stream API endpoint.
• Preferences:
  Pass --preferences key=value in CLI (repeatable) or in API payload.
• Get structured JSON:
  Use run-json in API mode or /output/json endpoint.

---

🔗 Key Documents & Resources

• ARCHITECTURE.md — Detailed architecture and flow diagrams
• PERSONA_DESIGN.md - Walkthrough deepdive of how personas are designed
• ./clients/README.md — CLI usage and switches
• ./apps/PERSONA — Persona YAML schema and examples
• ./apps/README.md — Persona/app design, best practices
• ./rasa/llm/LLM_CONFIG.md — LLM config and adapters
• CHANGELOG.md — Version history and recent changes

---

🛡️ Troubleshooting

• ModuleNotFoundError / import issues:
  Ensure you have __init__.py in rasa/, rasa/core/, clients/. Run from the project root.
• API not responding:
  Start with uvicorn rasa.api.main:app --reload and check port.
• Persona not found:
  Confirm apps/<persona>/persona.yaml exists and is valid YAML.
• LLM/backend errors:
  Check llm.py, try /llm/health or /status endpoints.

---

🤝 Contributing

• Fork, clone, and PR your features or fixes!
• Add new personas, frames, or operators following the standard structure.
• Write tests for all new logic—see tests/.
• Document YAML fields and code for onboarding.

---

RASA lets you build and operate real-world, context-aware, persona-aligned AI agents in your domain.
Get started, explore, and make it your own!