Trusted Execution Cluster (TEC) / Compute Layer
Overview
The Compute Layer is responsible for processing encrypted data securely using Trusted Execution (TE) nodes. These nodes run specialized compute programs on data without exposing the raw inputs. The Compute Layer seamlessly integrates with the rest of the system, ensuring that all computations are performed securely and efficiently.
TEE Nodes Structure
The Compute Layer is comprised of multiple TEE nodes that execute computation jobs in isolated and secure environments. Each of these nodes will run a software provide by Intract to ensure adherence to our guidelines and prevent any malicious behaviour. Also when a node joins the network, it is allocated a special NFT which works as an identity card of the network to prove to third parties (e.g. Lit Protocol) that it is part of the Intract Network. This could be redacted at any time in case of malicious behaviour.
When a job arrives to the TEE network for compute, all nodes which are up at that time are eligible to run it and it is allocated based on the amount staked as well as the reputation of the node. Everytime a node successfully completes a job, their reputation is increased and fees is allocated to them. However in case of any malicious reports, the reputation could be reduced and the stake could also be slashed.
Compute Job Input Structure
For each compute job, the following input components are provided:
Encrypted Data:
The data to be processed is encrypted with a designated key.
Decryption Method & Parameters:
Decryption Method: Specifies the method to decrypt the input data.
Decryption Parameters: Associated metadata required for decryption, which is itself encrypted using the TEE node's private key.
Example: The primary decryption method supported is the LIT protocol. For more details on the LIT protocol, please refer to the LIT Protocol Documentation.
Compute Program:
The actual program or script that needs to be executed on the decrypted data. This could be an ingestion module, a context model, or any other supported compute task.
Result Encryption Key:
The key that will be used to encrypt the final computation result before it is transmitted or stored.
Supported Decryption Methods
The Compute Layer supports multiple decryption methods to ensure flexibility and security. The primary decryption method is:
LIT Protocol:
Uses session keys to securely decrypt data.
The decryption parameters necessary for the LIT protocol are provided as part of the job input.
For comprehensive details on the LIT protocol, please refer to the LIT Protocol Documentation.
Note: Additional decryption methods may be supported in future updates. The protocol’s modular design allows for easy integration of new decryption techniques as needed.
Advanced Compute Methods for AI Agents
In addition to running traditional compute programs on decrypted data, our Compute Layer supports advanced AI-driven compute methods that transform raw data into specialized outputs. Depending on the nature of the input data and the specific requirements of the application, the secure compute environment can produce various structured outputs, such as:
Vector Database (VectorDB) Embeddings: For applications like information retrieval or semantic search, the compute node can convert raw data (e.g., emails, documents) into high-dimensional vector embeddings. These embeddings are then stored in a VectorDB, enabling fast and efficient similarity searches. For instance, if an AI agent queries travel-related information from a set of emails, the system can generate vector embeddings that capture the semantic essence of travel-related content, rather than exposing the entire email data.
Knowledge Graphs: When the context and relationships within the data are crucial, the compute node can generate a knowledge graph. This structured representation captures entities, relationships, and attributes from the input data, facilitating complex reasoning and insights. For example, a knowledge graph built from emails might reveal connections between contacts, topics, or events related to travel.
Retrieval-Augmented Generation (RAG): In scenarios where a condensed and query-specific representation is needed, the compute node may generate a smaller, focused Retrieval-Augmented Generation output. This involves summarizing or embedding only the relevant portions of the raw data, thereby optimizing the output for the AI agent's query while maintaining data privacy.
Workflow & Data Flow
Job Submission:
A compute job is initiated with the full input set (encrypted data, decryption method, metadata, result encryption key, and compute program).
Decryption Process:
The TE node retrieves the encrypted data and, using the specified decryption method (e.g., LIT protocol), decrypts the data with the provided parameters.
Execution of Compute Program:
Once decrypted, the compute program is executed on the data. The program may process, transform, or analyze the data as required.
Result Encryption:
After processing, the resulting data is encrypted using the provided result encryption key.
Result Delivery:
The encrypted result is then securely transmitted back to the requesting entity or stored as per system requirements.
Last updated