GURU AI Framework Offers Self-Evolving Control for Android

Open-source tool builds its own capabilities and manages your entire smart home.

GeoDetect PipelineNovember 16, 2026
Share

Overview

A new open-source project named GURU is challenging the conventional architecture of personal AI assistants by introducing a self-evolving framework for Android devices. Rather than functioning as a static chatbot or a simple task executor, GURU positions itself as a dynamic permission layer and cognitive engine that builds its own tools and automations over time.

The framework, developed under the banner of Unus Lumen, claims to offer a level of agency previously unseen in consumer AI. It does not merely describe actions users could take; it executes them across the user’s entire digital and physical network. The system is designed to become more useful the longer it is owned, learning from interactions and expanding its capabilities without requiring external software updates.

"It is not a chat app. It is a permission layer across 167 Android permissions, a tool registry of around 390 tools and growing, an encrypted memory engine written in Rust, and a full Unix userland bundled inside the APK."

Technical Architecture and Memory

At the core of GURU is a cognitive architecture built in Rust, designed to operate entirely within the device’s local environment. This ensures there is no external server in the data path and no open ports exposed to the internet, addressing significant privacy concerns inherent in cloud-dependent AI models. The engine runs in-process, sealed within an encrypted vault that uses a passphrase generated during the initial setup.

Unlike traditional assistants that reset their state with every session, GURU maintains a persistent, evolving memory. This memory is structured into three distinct layers:

  1. BUFFER: A temporary holding area where new facts and interactions initially enter.
  2. EPISODIC: Memories are promoted here once they are recalled sufficiently, preserving specific events and timelines.
  3. SEMANTIC: Long-term knowledge that settles here after repeated reinforcement, forming the base of the system’s understanding.

The system employs a graph-based model to record relationships between memories, including temporal, causal, and contradictory links. A daily consolidation pass merges duplicates and prunes weak associations, while a seven-day exponential decay half-life ensures outdated information fades naturally. Notably, the framework avoids true deletion; instead, it archives memories to maintain a rewritable trail, allowing the AI to reference conversations from months prior without prompting.

Network Coverage and Protocol Implementation

GURU differentiates itself by speaking the underlying protocols of devices rather than relying on proprietary partnerships with specific brands. This approach allows for broad compatibility across smart home ecosystems, gaming consoles, and legacy hardware. The framework’s tool registry currently encompasses approximately 390 tools, covering a wide array of interaction methods:

  • UPnP and SOAP: Used for devices that expose services, enabling direct control of networked hardware.
  • DIAL: Facilitates casting and application launching on televisions, with logic to distinguish between devices that refuse commands and those that are simply offline.
  • Wake-on-LAN: Allows GURU to rouse sleeping devices before issuing further instructions.
  • Android TV Remote: Implements the native protobuf protocol to control devices from manufacturers such as Sony, Philips, and Nvidia Shield.
  • Console Protocols: Includes Xbox SmartGlass support, handling complex key exchanges and AES-CBC encryption channels required by gaming hardware.
  • SNMP and SSH: Provides access to enterprise-grade networking equipment and any machine with a shell interface.
  • IR and ADB: Utilizes Broadlink blasters or in-built phone emitters for infrared control, and Android Debug Bridge for direct device management.

Verification is a critical component of this control layer. GURU employs a write-and-read verification cycle for every command. After sending a command, it reads the device’s current state and compares it against the expected outcome. If a discrepancy exists, or if a device reports a capability it does not actually possess, the system refuses the action and logs a detailed reason. This prevents silent failures, ensuring that users receive accurate feedback on whether an action succeeded or failed.

Discovery and Security Philosophy

The framework’s discovery engine is designed to provide an honest representation of the user’s network. Traditional network scanners often rely on curated port lists or broadcast responses, which can miss devices that do not advertise themselves openly. GURU sweeps every port and properly reads mDNS and SSDP traffic, arranging multicast locks to ensure comprehensive coverage.

Endpoint fingerprinting is conducted through HTTP responses, certificate analysis, DIAL descriptions, and UPnP device documents. This allows the system to identify what a device actually is, rather than just recording its IP address. The framework explicitly distinguishes between a refusal to connect and an absence of a device, preventing false positives in network maps.

Security is approached through the principle of least privilege and local execution. By bundling a full Unix userland inside the Android application package, GURU operates with the same isolation constraints as other apps but with significantly expanded capabilities within that sandbox. The memory engine’s encrypted vault ensures that sensitive personal data and automation histories remain inaccessible to external parties, reinforcing the claim that this is a tool for local autonomy rather than data extraction.

Implications for Personal AI

The release of GURU represents a shift in how developers and users conceptualize artificial intelligence on mobile platforms. By moving away from the “assistant” paradigm toward a “capability builder” model, the framework invites users to treat their phones as central hubs for complex automation rather than simple communication devices.

The self-evolving nature of the system suggests a future where AI tools can adapt to niche or unusual hardware setups without waiting for vendor support. As GURU encounters new devices or protocols, it writes its own drivers and skills, effectively expanding its own user manual. This architecture challenges the status quo of closed ecosystem lock-in, offering a path toward truly interoperable smart home environments controlled by a single, intelligent interface.

While the project remains in active development, its emphasis on verifiable control, persistent memory, and local security sets a new benchmark for what personal AI frameworks can achieve. As the community of contributors grows, the tool registry and protocol support are expected to continue widening, potentially redefining the relationship between users and their digital infrastructure.

Share

Sources

Tools mentioned