Intentus - Omni Robotic Orchestrator

Define the next generation of human-robot interaction. Built by PerceptusLabs

Intentus is a scalable, self-learning orchestration server that brings audio-visual intelligence to robots and autonomous agents. With memory, modular tools, and a structured planning system, it can perceive, reason, and act—adapting intelligently with each interaction.

“The future isn’t just about robots that respond — it’s about robots that understand, plan, and grow.”

---

✨ Key Capabilities

• Vision-Language-Action (VLA) Pipeline Send structured text representations of audio/visual input; Intentus interprets, plans, executes, and adapts.

• Web-Enabled Agent Includes a web-browsing toolchain for open-ended question answering, research, or data retrieval.

• Tool-Expandable Add custom robot skills via natural language in under 100 lines. Tools live in /tools and can be hot-loaded.

• Memory-Backed Planning Multi-step agents with persistent memory and feedback-driven self-improvement.

• Production-Ready API-based architecture, secure with API key support, logging, and environment configuration.

• Asynchronous & Scalable Modern Python backend, fast async processing, with customizable agent parameters and execution lifecycles.

---

📦 Installation

git clone https://github.com/Perceptus-Labs/Intentus.git
python -m venv venv
source venv/bin/activate
pip install -e .

---

🚀 Quickstart: Local Demo

python intentus/examples/agent_demo.py

Demo Task: “What is the capital of France?”

You’ll see:

• Multi-step reasoning
• Query analysis
• Final output
• Agent memory dump
• Execution time and path

💡 See real-time agent logs in example.log, outputs saved to example_outputs/.

---

🧠 How It Works

1. Your robot/system sends a structured intention (transcribed speech, vision cues, context, etc.)
2. Intentus parses & analyzes the command with optional tools
3. It plans actions, reasons over them, and executes tools
4. Feedback updates its memory, and it continues iterating
5. Returns a full report on results, reasoning, and future suggestions

---

🛠️ Add Your Own Tools (VLA Plugins)

Developers can extend robot abilities via natural language definitions and lightweight Python handlers:

# Create a new tool file in intentus/tools/
# Describe your tool's purpose in plain English
# Add a function with input/output contracts

Your robot can learn to:

• Navigate new terrain
• Run factory checks
• Diagnose mechanical issues
• Interface with APIs or microcontrollers

🔧 Tool architecture is fully modular and scalable.

---

🧪 Run the API Server (Orchestrator)

The orchestrator receives intention payloads from your robot/Go backend:

1. Install requirements

pip install -r requirements.txt

2. Set environment variables (either in .env or directly in the code)

export ORCHESTRATOR_API_KEY="your-api-key"
export ORCHESTRATOR_HOST="0.0.0.0"
export ORCHESTRATOR_PORT="8000"

3. Run the server

python main.py

Now available at: http://localhost:8000

---

🔍 API Reference

POST /orchestrate

Receives a structured intention payload:

{
  "session_id": "session-123",
  "intention_type": "user_query",
  "description": "User asked about the weather",
  "confidence": 0.95,
  "transcript": "What's the weather like today?",
  "environment_context": "User is in San Francisco, CA",
  "timestamp": 1703123456
}

Returns:

{
  "success": true,
  "query_analysis": "...",
  "base_response": "...",
  "final_output": "...",
  "execution_time": 2.5,
  "steps_taken": 3,
  "memory": ["Step 1: ...", "Step 2: ..."]
}

Authentication: Authorization: Bearer your-api-key

---

🔍 Example Use Case

Imagine a robot assistant in a kitchen:

1. Sees an image of spilled flour (visual cue)

2. Hears “What do I do now?” (audio command)

3. Intentus processes the environment:

   • Identifies the spill
   • Suggests cleanup plan
   • Executes cleaning tool
   • Stores feedback if plan failed/succeeded

This is closed-loop intention orchestration — and it’s just the beginning.

---

🧰 Developer Configuration

Customize the agent in main.py > get_agent():

AgentConfig(
    llm_engine="gpt-4.1-mini",
    enabled_tools=["Wikipedia_Knowledge_Searcher_Tool"],
    verbose=True,
    max_steps=5,
    temperature=0.7,
)

---

✅ Testing

python test_orchestrator.py

Make sure API_KEY in the test script matches your environment variable.

---

🔐 Security

• /orchestrate is protected by an API key
• Set ORCHESTRATOR_API_KEY
• Authentication is optional but highly recommended

---

📈 For Investors & Partners

Intentus is a foundational system in robot cognition:

• It’s tool-agnostic and can support a fleet of heterogeneous robots
• It bridges perception, language, and control
• It allows natural language tool creation — no low-level firmware work required
• It logs full memory and reasoning chains, giving you total auditability
• It is modular, API-first, and cloud-compatible

Let’s redefine how robots understand and respond to the world.

---

👥 About PerceptusLabs

We are building the future of robotic cognition — creating agents that not only follow instructions, but think, adapt, and evolve.

Visit perceptuslabs.com for more.