VERITAS OS is a cutting-edge Proto-AGI framework designed to elevate large language models (LLMs) beyond traditional chatbots, transforming them into a robust, reliable, and auditable decision-making operating system. This advanced framework leverages the structured decision pipeline encapsulated in a single /v1/decide API call, which meticulously guides LLMs through a deterministic process that includes Evidence, Critique, Debate, Planner, ValueCore, FUJI, and TrustLog.

Key Features

• Integrated Decision Framework: The platform functions as a complete Decision/Agent OS, infusing safety, memory, value functions, and world-state management into the operation of LLMs as reasoning engines.
• Structured Outputs: Each call to /v1/decide returns a detailed JSON payload with insights such as:
  • chosen: The selected action along with its rationale and uncertainty levels.
  • alternatives[]: An array of other options considered during the decision-making process.
  • evidence[]: References to the evidence utilized in formulating the decision.
  • critique[]: Self-critiques performed on the candidate plan.
  • debate[]: Multi-perspective internal debates that assess different viewpoints on the decision.
  • telos_score: Indicator of alignment with the current value function.
  • fuji: A safety and ethics assessment status (allow, modify, block, or abstain).
  • trust_log: A hash-chained log of entries for auditing purposes.

API Interaction

VERITAS OS is exposed as a local FastAPI service that can be accessed via OpenAPI 3.1, enabling seamless interaction through tools like Swagger Editor/Studio. A valid X-API-Key is required for all endpoints to ensure the integrity and security of interactions with the API.

Contextual Decision Making

The framework supports AGI-style tasks by utilizing a contextual schema formeta-decisions. A Context object encapsulates user queries, goals, constraints, and preferences, allowing users to leverage the system to make high-level decisions for AGI research, self-improvement cycles, or experimental plans demonstrating rigorous safety adherence.

Core Architecture

The dizzying array of features of VERITAS OS derives from its modular architecture. The following components are noteworthy:

• Core OS Layer: Manages overall orchestration, pipeline execution, planning, and decision outputs.
• Safety and Ethics: Implements rigorous safety measures and compliance checks via the FUJI Gate, assessing decisions against ethical standards.
• MemoryOS: Manages long-term memory storage and retrieval, ensuring relevant historical data informs current decisions.
• Logging and Auditing: Ensures comprehensive logging of decisions and actions to maintain trust and transparency throughout the decision-making process.

Conclusion

VERITAS OS stands as a pioneering framework aimed at researchers and developers interested in AGI and AI safety explorations. By facilitating safe, repeatable, and auditable decision-making processes, it lays the groundwork for robust autonomous agents capable of self-improvement and ethical operation.