Apollo-1: From Generative AI to Controllable AI

Generative AI ignited the first wave of conversational AI, yet it speaks only for the individual user. When an agent must act on behalf of an entity—an airline, retailer, or bank—stochastic output becomes an existential liability: broken compliance, policy violations, no audit trail. To reach the necessary accuracy, trust, and reliability, the core must shift from neural to neuro-symbolic reasoning.

The Second Wave: Controllable AI

The second wave is Controllable AI, which unlocks conversational agents that operate on behalf of the entity that deployed them, rather than the end-user, producing controllable outputs while maintaining the language capabilities of generative AI. Controllable outputs let companies guide agents toward desired behaviors whenever they choose.

From LLMs to Neuro-Symbolic Models

Apollo‑1 is our neuro-symbolic foundation model for conversational agents. In Apollo-1, a Symbolic Reasoner replaces the transformer as the model’s decision-making core. 

User input is parsed by lightweight NLP modules and translated into a symbolic language of controllable entities (e.g. intent, constraints, context). The Symbolic Reasoner uses these entities to plan and trigger external actions. Action responses pass back through the same NLP modules, are re-encoded as controllable entities, and inform the Symbolic Reasoner, which then either produces the final model output or keeps invoking additional actions and tools until the interaction is complete.

How Apollo-1 Enables Controllable Interactions

• Traceable execution — provides a white-box view of its symbolic reasoning: every sub-action is logged, inspectable, and editable, replacing the black-box opacity of LLMs.
• Real-time Control Panel — enables operators to inspect reasoning, adjust memory schemas, inject rules and replay live trajectories.
• Deterministic control on demand — lets operators inject rule-based logic whenever needed. 
• Continuous fine-tuning — Apollo-1 learns from every feedback; teams can fine-tune and test live for continuous improvement.

Why Controllable Outputs Beat Purely Generative Outputs

• Controllable interactions enable more complex tool use, deliver higher performance, and unlock new use cases.
• Symbolic policy engines validate compliance, tone, and brand alignment in real time.
• Ground-truth connectors call live APIs for inventory, availability, user details, etc.—never hallucinated or cached text.
• Full audit trail — every step is logged and reviewable (mandatory for regulated industries).
• Generative fluency preserved — retains full natural-language prowess of top LLMs.

Hard Proof Across the Toughest Tests

* Final Pass^¹, Pass^², Pass^³, and Pass^⁴ numbers will be published in our official τ-Bench report in July 2025. In the Gemini Flights and Amazon Rufus tests, Apollo-1 taps the same live APIs as its generative rivals.

Bottom Line

For nearly eight years we pursued one question: how to build conversational agents that act for an airline, bank, or university—not merely assist a user. Generative models break down on compliance, auditability, and deterministic control. Apollo-1 solves this—combining generative fluency with fully controllable, policy-safe interactions. Running at scale inside Fortune 500 companies; strategic go-to-market partnership with Google; General Availability in September 2025.