Tencent Cloud officially releases the cloud-based intelligent body CodeBuddy NPC.

On July 23rd, Tencent Cloud officially released the cloud intelligent entity CodeBuddy NPC. As a new generation intelligent entity launched by Tencent Cloud for enterprise R&D processes, NPC relies on the Tencent Cloud R&D platform CNB to directly utilize code repositories, issues, pull requests, CI/CD, and other R&D contexts to continuously advance tasks in the existing R&D process, eliminating the need for developers to repeatedly copy requirements, code, and error logs. Developers only need to assign a task, and CodeBuddy NPC can autonomously complete solution planning, code development, submit PRs, perform tests, and continuously modify based on CI results until delivering acceptable results. A batch of ready-to-use official R&D NPCs are waiting to be summoned in CNB, supporting multiple NPC team formations to collaborate on complex R&D tasks. The first-round token consumption of official R&D NPCs has been reduced from over 20,000 tokens to about 2,000 after continuous optimization, a decrease of over 90%. NPC possesses R&D memory, collaborating like a colleague Unlike AI programming tools mainly running locally and aiding in code writing, CodeBuddy NPC no longer stays within the code editor but directly enters the R&D process, continuously working in code repositories, issues, PRs, CI/CD, and other stages. Developers can directly address the NPC like a colleague in issues. Once a task is received, it will automatically read various contexts in the R&D process: - Issues are requirement memory, documenting tasks, backgrounds, and objectives; - Code repositories are project memory, containing project structures, dependencies, and evolution history; - PRs are change memory, depositing solutions, codes, and review feedback; - CI is quality memory, recording build results, failure reasons, and repair processes. These R&D assets are no longer just the developers' work records but also form the native working context of NPCs. With this "R&D memory," NPCs can continuously complete R&D tasks rather than generating code all at once. Development Website Automatically read requirements and project code, complete development, submit PRs, and generate preview environments. Autonomously resolve code conflicts Automatically handle code merge conflicts, modify and resubmit. Automatically fix CI Automatically identify and fix code issues after CI failures, re-execute until passing quality gates. Form NPC Teams to complete complex R&D tasks In addition to independent task completion, CodeBuddy NPC supports multiple NPCs working together in different roles. We used an NPC Team with zero human intervention to complete the full development process of a game from requirement decomposition to final validation. The project manager NPC is responsible for requirement decomposition and task assignment; scriptwriters, artists, composers, voice actors, and other NPCs complete content creation individually; asset integration NPCs consolidate code and materials; code review NPCs and test NPCs perform final validation. Enterprises can also arrange their own NPC Teams through roles, skills, SOP, and Team, allowing NPCs from different roles such as product, development, and testing to collaborate and complete complex R&D tasks, such as defining a "yuanfang." Reliable work without token expenditure, enterprises can flexibly adjust the model Executing a task once usually requires multiple rounds of model calls. As the first-round system prompts and tool descriptions are continuously reused in subsequent calls, the smaller the first-round token consumption, the lower the overall cost of multiple rounds. Tencent Cloud continuously optimizes prompts, tool calls, CLI outputs, and cache hit rates, reducing the first-round token consumption of official CodeBuddy NPCs from over 20,000 to about 2,000, a reduction of over 90%, significantly reducing the overall token expenditure for complex R&D tasks. Enterprises can also flexibly configure model strategies based on task complexity. Simple tasks can call low-cost models, complex R&D tasks can utilize more powerful models, and multi-modal tasks can run corresponding models in the cloud to achieve a balance between performance and cost. Relying on the Tencent Cloud R&D platform CNB, enterprises do not need to overhaul their existing toolchains and can continue to use Git, build, deploy, TAPD, and other R&D tools, directly assigning tasks to NPCs within their original processes.