BRAWL Agent v3.1

Autonomous Intelligence. Relentlessly Capable.

BRAWL is a high-performance, autonomous AI agent CLI powered by Ollama, smolagents, and RTK (Recursive Token context). Built for power users who need persistent memory, advanced toolsets, and multi-provider flexibility.

---

Quick Start

1. Clone & install dependencies

pip install -r requirements.txt

2. Configure environment

Create a .env file (see .env.example for details):

OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=qwen3-coder-next:cloud
PROVIDER=ollama
MAX_CONTEXT_TOKENS=4096

3. Launch BRAWL

python main.py

---

Features

• Multi-provider support: Seamlessly switch between Ollama, HuggingFace, and NVIDIA models.
• Smart context (RTK): Intelligent token management with recursive summarization to maintain long-term context without hitting limits.
• Persistent memory: Chat sessions are automatically saved and can be resumed at any time.
• 120+ tool suites:
  • file_tools, search_tools, code_tools for system automation.
  • web_tools, network_tools, ai_tools for connectivity and analysis.
  • database_tools, data_tools, math_tools for structured data handling.
• Interactive REPL: A polished command-line interface with built-in session management. ...more

---

CLI commands

Command	Action
/new	Start a fresh chat session
/chats	List all saved sessions
/switch <id>	Switch to a specific session
/delete	Delete the active (or specified) chat
/model	Manage providers and models (list, use, delete)
/clear	Refresh the terminal view
/help	Display the command manual
/exit	Terminate the BRAWL session

---

Project structure

• main.py: Entry point for the BRAWL CLI.
• core/: Core logic (Agent management, RTK memory, configuration).
• tools/: Modular tool implementations (filesystem, web, database, etc.).
• chat_storage/: Persistent SQLite storage for sessions.
• output/: Default directory for generated files and exports.

---

Credits

Credits to smolagents , rtk-ai and Ollama.