Comate (Wenxin Kuaima)

Type: full-code · Vendor: Baidu (文心快码) · Language: TypeScript (extension) · License: proprietary (free individual + enterprise) · Status: active · Status in practice: mature

Links: homepage docs

Baidu's coding assistant on Wenxin (Ernie) models, providing IDE-integrated code completion, generation, and chat across Baidu's own programming corpus and external open-source data.

Description. Comate (文心快码, formerly Baidu Comate) is Baidu's enterprise-grade coding assistant powered by Wenxin (Ernie) large models. It ships as a VS Code / JetBrains extension and offers completion, function generation, code explanation, test generation, and a chat mode with repository context. It positions itself for Chinese-language software teams and ships an enterprise edition with private deployment, fine-tuning on the customer's codebase, and integration with Baidu's broader Qianfan platform.

Agent loop shape. IDE extension calling hosted or in-VPC Wenxin models. Completion is single-shot per cursor event; chat runs short tool-loop with file context. Enterprise edition supports private RAG over the company's codebase.

Primary use cases

code completion and chat in VS Code / JetBrains for Chinese software teams
enterprise private deployment with code-base fine-tuning
integration with Baidu Qianfan platform for end-to-end MLOps
Chinese-language natural-language → code generation

flowchart TD USER[Developer] --> EXT[Comate extension<br/>VSCode / JetBrains] EXT --> DEPLOY{Deployment} DEPLOY -->|public| WENXIN[Wenxin/Ernie<br/>hosted] DEPLOY -->|enterprise| PRIV[In-VPC Wenxin<br/>+ private RAG] PRIV --> REPO[(Company codebase<br/>vector index)] REPO -->|context| PRIV WENXIN --> EXT PRIV --> EXT EXT --> USER

Key concepts

Wenxin (Ernie) backing — Coding-tuned variant of Baidu's Ernie model family powers the assistant.
Enterprise private deployment — In-VPC inference with optional fine-tuning on the customer's repository.
Codebase RAG (enterprise) → naive-rag — In-VPC retrieval over the customer's codebase for repository-aware completion.
Qianfan platform integration — Plugs into Baidu's broader MLOps stack rather than standing alone.

Patterns this full-code implements —