Over the past year, various online platforms have launched character Agent interactive products one after another, with film and TV character dialogues, virtual entertainers, and interactive drama agents landing intensively. However, in actual applications, there are widespread problems such as weak character fidelity, distorted personas, answers drifting away from the setting after multiple rounds of conversation, and sluggish experiences under multiple users and high concurrency. How can reliable infrastructure support ensure that character Agents interact smoothly while keeping the persona intact?
On October 25, the 2026 Three-Body Day·Dehydration-Soaking Carnival was held in Beijing. During the event, Wuxin Qiong and Three-Body Universe officially announced a strategic partnership, with Wuxin Qiong co-founder and CEO Xia Lixue and Three-Body Universe CEO Zhao Jilong jointly releasing the first outcome of their collaboration — an AI dialogue Agent with Sophon personality traits, "Sophon Agent 1.0".
As Three-Body Universe's AI infrastructure partner, Wuxin Qiong provides full-chain engineering capability support for intelligent agent development, knowledge and memory management, and inference services through its Agentic Infra product and service system, supporting Three-Body Universe in building highly polished role-playing Agents.
First collaboration outcome lands: building a character Agent centered on multi-agent collaboration
In the narrative of The Three-Body Problem, Sophon is an important character connecting Trisolaran civilization and the human world. The Administrator Sophon in Miniature Universe 637 is a "copy of Sophon" and assumes the actual management duties of the miniature universe. Sophon Agent 1.0 chose the Administrator in Miniature Universe 637 as its character anchor — an identity that provides the agent with a distinct character foundation while also placing clear requirements on its mode of expression and the boundaries of its answers.
The product takes character consistency and interaction continuity as its core design focus, deeply integrating the character's expression style, worldview knowledge system, and conversational memory capabilities. Users can experience multi-round conversations with Sophon, perceiving Sophon's linguistic features and character temperament during the exchange, while conversational memory can persist across sessions, enabling coherent long-term interaction. In the future, as knowledge organization, memory management, multi-agent collaboration, and multimodal interaction capabilities continue to accumulate, the Sophon Agent will keep growing from the Administrator identity, gradually enriching the IP personality and evolving toward a more complete "Sophon" full form.
In this collaboration, Three-Body Universe is responsible for worldview settings, character boundaries, and personality trait fidelity; Wuxin Qiong, as the AI infrastructure partner, provides support with its Agentic Infra product and service system, offering Three-Body Universe full-chain engineering capabilities including intelligent agent development, knowledge and memory management, and inference services. For IP character interaction scenarios, Wuxin Qiong centers on multi-agent collaboration to support Three-Body Universe in building character Agents and underlying service infrastructure, enabling it to incorporate the generative expressiveness of role-playing and the precision of worldview knowledge into a unified system design, and through collaborative mechanisms helping Three-Body Universe balance personality expression, narrative consistency, and knowledge constraints.
Wuxin Qiong serves Three-Body Universe with Agentic Infra, supporting the resolution of three major engineering pain points of character Agents
The implementation of Sophon Agent 1.0 relies primarily on the full-chain technical capabilities of Wuxin Qiong's Agentic Infra, addressing industry-common pain points of IP character Agents and providing Three-Body Universe with systematic capability support spanning development platforms, operating mechanisms, and underlying architecture.
AgentWorks platform: enabling Three-Body Universe to turn "character fidelity" from prompt engineering into a configurable systems engineering effort
The construction of character Agents depends on high-quality worldview text resources. With traditional prompt engineering, it is difficult to guarantee the accuracy of knowledge retrieval and the fidelity of the character's persona.
Based on Wuxin Qiong's AgentWorks intelligent agent development service platform, Three-Body Universe can start from three levels — knowledge management, character expression, and interaction memory — to transform The Three-Body Problem worldview from text materials into configuration resources that the Agent can invoke, trace, and reuse at runtime, allowing Sophon Agent 1.0 to strictly constrain knowledge boundaries while maintaining personality authenticity.
- At the knowledge management level, the platform's two-layer "Collection–knowledge base group" architecture supports Three-Body Universe in ingesting multi-format worldview materials, automatically completing preprocessing such as parsing, chunking, and embedding; through unified retrieval parameter configuration at the knowledge base group level, it supports centralized maintenance of worldview materials and their reuse across multiple agents.
- At the character expression level, structured, multi-layer prompts manage the character's personality, tone, and action norms, supporting dynamic adjustment by dialogue scenario to ensure persona consistency across different contexts.
- At the interaction memory level, relying on the platform's self-developed memory capability Infini Memory, it helps Three-Body Universe solve problems of traditional memory systems such as fragmentation, factual conflicts, and difficulty updating knowledge, achieving cross-session memory persistence and precise retrieval.
Multi-agent collaboration mechanism: helping Three-Body Universe break down "no persona collapse" into verifiable quality metrics
A common industry problem with IP role-playing is that "knowledge density" and "persona tension" are hard to balance: directly stuffing large amounts of retrieved settings into prompts easily causes the model to degrade into a stiff "settings manual", losing the character's appeal; overly liberalizing generation freedom easily leads to persona drift and worldview "breakdowns".
To address this, Wuxin Qiong implemented pipeline isolation of character expression, knowledge retrieval, and content verification in the operating chain, forming a complete "generate–discriminate–targeted correction" closed loop through multi-agent collaboration.
In a single conversation, persona-styled generation and worldview knowledge retrieval proceed in parallel, preventing retrieved text from directly polluting the character's tone and manner; after the draft is generated, an independent review module specifically checks for two types of problems — persona OOC and worldview setting violations — and content that fails verification is returned with specific conflict diagnostics for revision.
Meanwhile, Wuxin Qiong also established a positive/negative example evaluation system of "automatic model verification plus human review": through Trace chain tracing of the causal relationships of inference and tool calls, content hallucinations are investigated at the source; automated modules compare generated results sentence by sentence against setting sources to locate factual deviations. Deterministic "breakdown and dissonance" problems are automatically intercepted and corrected by the system, while higher-dimensional character "dramatic sense and spirit" is reviewed by the Three-Body Universe content team, balancing implementation efficiency and character quality.
PDD cross-cluster heterogeneous inference: ensuring a low-latency, low-cost character experience for Three-Body Universe
Compared with ordinary dialogue Agents, character Agents generally feature more dialogue turns, longer contexts, and higher costs of maintaining personality consistency; multi-round conversations also accumulate long contexts, and knowledge retrieval, memory invocation, and content verification add chain complexity, plus the high-concurrency demands brought by multi-user interaction, placing higher requirements on underlying inference capabilities.
Wuxin Qiong adopted its self-developed PDD cross-cluster heterogeneous inference architecture, deconstructing the traditional "P-D" chain into a three-stage separated inference architecture of "P-RLD-MD". Leveraging the extremely high prefix cache hit rate characteristic of Agent business, it deploys RadixCache prefix caching on the Decode side, reducing the data volume of KV Cache transmission across clusters by an order of magnitude, bypassing the bandwidth and latency bottlenecks of KV Cache transmission in Ethernet environments. Measured data shows this architecture can reduce first-Token latency by 51.5% and per-Token cost by 37.5%, providing technical support for Three-Body Universe to safeguard the Sophon Agent experience in long-context and concurrent interaction scenarios.
Accumulating reusable capabilities to serve Three-Body Universe in exploring an industrialization path for IP character Agents
For a long time, the field of IP character agents has had a structural contradiction: technical teams are mired in underlying engineering details while content teams struggle to participate deeply in product polishing. What Sophon Agent 1.0 validates is not just the technical feasibility of a single character, but an engineering path for IP character agents to move from project-based work toward assetization.
Based on Wuxin Qiong's AgentWorks platform, Three-Body Universe can encapsulate the engineering complexity of the full agent lifecycle within the platform, focusing more on worldview design and experience polishing without repeated investment in underlying engineering construction. Meanwhile, the platform has built-in complete reusable capabilities, supporting the reuse of agent templates, knowledge resources, and capability configurations, with memory resources manageable independently according to application needs; at the service access level, it supports API integration with business systems, so future expansion of new characters and new scenarios can reuse existing engineering capabilities, helping Three-Body Universe reduce marginal implementation costs.
The release of Sophon Agent 1.0 is both an exploration by Three-Body Universe of extending its character-building and cross-media content development capabilities into the AI interaction field, and a practice by Wuxin Qiong of supporting Three-Body Universe with its Agentic Infra product and service system to advance the implementation of agent technology and engineering services in sci-fi IP scenarios. From concept planning to production-grade product implementation, it has validated Wuxin Qiong's Agentic Infra delivery capability serving Three-Body Universe's IP character scenarios, and provided a reusable paradigm for sci-fi IP to extend from content assets into interactive digital assets.
In the future, the two parties will further explore directions such as character long-term memory and relationship evolution, multi-character shared-worldview collaborative narrative, and multimodal cross-media interaction. Wuxin Qiong will also continue to accumulate common infrastructure capabilities supporting multi-IP, multi-character, and multi-terminal operations, providing reusable underlying support for IP holders including Three-Body Universe, and driving sci-fi IP AI Agents from individual cases toward industrial development.
This article was provided by Wuxin Qiong and reprinted by QbitAI with authorization; the views belong to the original author.