Status: Archived for now. I may return later to continue the Nexara docs and development.
Production-grade autonomous AI agent framework by NexaraAI and DemonZ Development
Nexara V1.0 is a modular autonomous agent framework built for resilient execution across multiple environments.
It combines:
- Multi-provider LLM failover
- Semantic long-term memory
- Dynamic skill loading
- Autonomous task replanning
- Cross-platform execution awareness
Designed to operate consistently across mobile and desktop environments while loading only the capabilities supported by the current host system.
Most autonomous agent frameworks fail in real-world deployment because they assume:
- one provider will always stay online
- every system exposes identical tools
- failures should terminate execution
Nexara was built to solve that.
It routes intelligently across providers, isolates execution, and replans when skills fail instead of stopping.
User Message
│
▼
┌─────────────────────────────────────────────────────┐
│ LLM Router │
│ Gemini → Groq → Ollama → OpenAI / NVIDIA │
│ Automatic failover with backoff │
└──────────────────────┬──────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────┐
│ ReAct Loop │
│ THINK → ACT → OBSERVE → REPLAN (up to 14 cycles) │
│ Automatic retry on skill failure (3x) │
└──────────────────────┬──────────────────────────────┘
│
┌─────────────┼─────────────┐
▼ ▼ ▼
Skills Memory Planner
Dynamic Load SQLite + Vectors Background Tasks
Nexara pulls skills dynamically from the official skill warehouse and activates only host-compatible modules.
- Native device control
- SMS access
- Camera integration
- Battery telemetry
- Call log access
- APT package management
- systemd service control
- Docker orchestration
- cron automation
- PowerShell execution
- Registry inspection
- Event Log monitoring
- Homebrew integration
- AppleScript execution
- Spotlight search
- Web scraping
- File operations
- Python execution
- SQLite queries
- REST webhooks
Current warehouse size: 86+ dynamically loadable skills
| Feature | Beta | Nexara V1.0 |
|---|---|---|
| Tool Calls | Regex JSON parsing | Native structured function calling |
| LLM Engine | Single provider | 4-tier fallback chain |
| Memory | SQLite LIKE queries | Semantic vector similarity |
| Execution | Direct exec() |
Subprocess isolation + AST scanning |
| Autonomy | Stops on error | ReAct replanning loop |
| Scheduling | None | Natural language scheduler |
- Host-aware execution
- Failure tolerance
- Minimal resource overhead
- Expandable architecture
- Autonomous recovery
| Command | Access | Description |
|---|---|---|
/start |
All | Wake agent and sync skills |
/memory [query] |
All | Search semantic memory |
/run <goal> |
Admin | Queue autonomous task |
/tasks |
Admin | View active tasks |
/monitors |
Admin | View condition monitors |
/llm |
Admin | Check provider health |
/stats |
Admin | Full host system snapshot |
git clone [https://github.com/NexaraAI/nexara-agent.git](https://github.com/NexaraAI/nexara-agent.git)
cd nexara-agent
pip install -r requirements.txt
chmod +x start.sh
./start.sh
---
## 🧩 Project Structure
```text
nexara-agent/
├── agent/ # ReAct engine, semantic memory, LLM router
├── utils/ # Dynamic skill_loader and system helpers
├── main.py # Execution logic
├── start.sh # Automated launch script
├── requirements.txt # Dependencies
└── .env # API keys & configuration
---
## 🔭 Roadmap
- Advanced multi-agent orchestration
- Distributed skill execution
- Expanded semantic memory indexing
- Native dashboard interface
---
## ⚖️ License
Licensed under **Apache License 2.0**.
See `LICENSE` for full terms.