Architecture

Onchain Computers are highly capable systems built upon composable patterns.

At their core lies an augmented LLM equipped with an adaptive persona generator, that allows the system to utilize revised system prompts for each interaction while maintaining a consistent character. Consequently, the model can exhibit parallel, evolving personalities tailored to specific activities—much like how a person behaves differently at work on a Monday compared to the Christmas table with family.

The core further contains an orchestrator engine that acts as the central controller, responsible for managing specialized workers with well-defined capabilities. (e.g., content delivery, user verification, using onchain wallets)

Worker LLMs can utilize chain of thoughts for perfecting an answer and multi-agent communication for handling complex tasks. The Call Chain is a flexible concept, built on common patterns that users can shape and combine to suit various use cases. Complexity is only introduced when it demonstrably improves outcomes.

Once all delegated workers have completed their tasks, a Synthesizer aggregates their responses to formulate the final output. While all intermediary LLM calls are stored and monitored, only these synthesized answers are permitted to exit through one of the interfaces. These responses are then:

• Sent as messages back to users on platforms like X (formerly Twitter) or Telegram.

• Reported to the governance module.

Additionally, they can operate intrinsically, governed by their Job Scheduler—a component of the orchestrator module that defines their will, responsibilities, and customs.

Long-Term Memory enables Onchain Computer to remember conversations with every individual they have interacted with, and thanks to the identity integrations, these conversations are being associated with each other over time when users verify themselves on multiple platforms. Regardless of the platform used, the Onchain Computer retains memory of previous interactions.

Monitoring is a critical component for:

• Self-Healing: Automatically forgetting irrelevant information or fixing issues autonomously.

• Memory and Knowledge Management: Ensuring the system remains efficient and up-to-date.

Over time, Onchain Computers autonomously manage tasks such as:

• Forgetting outdated or unnecessary information.

• Repairing any broken functionalities.

Smart Contracts play an important role of establishing an ownership and governance layer for each Onchain Computer, and allow verified human intervention, when necessary.

The Knowledge Module acts as a distinct "library" for both the agent and worker LLMs. This is the place where all the relevant onchain transactions are indexed, forums and social media platforms scraped, and non-LLM workers continuously organizing or transforming data to achieve certain goals, such as reducing redundancy, improving data integrity and making everything searchable. This "library" can be shared between Onchain Computers, reducing the time and cost before a new agent can start operating.

Now here is the full Architecture: