Miaowen AI Marketing Assistant

The Problem It Solves

Digital marketing teams often face challenges such as inefficient data queries, lack of creative inspiration, high ad rejection rates, and complex campaign operations. For example, retrieving daily account reports requires manual platform login and data export; creative generation lacks industry references; and post-rejection analysis is time-consuming to pinpoint and fix issues, impacting marketing effectiveness.

Core Capabilities and Workflow

Tencent Marketing Miaowen, based on Tencent Ads' "Miaowen" platform open API, provides two service modes:

• Chat Mode (primary): Handles intent recognition, tool dispatch, and reasoning via natural language interaction. Suitable for knowledge Q&A, data queries (e.g., daily/hourly reports, audience profiles), account analysis, ad diagnosis, creative inspiration (filtered by industry/placement), material review (pre-review/rejection analysis), and campaign operations (creating/modifying ads).
• API Mode (lightweight direct query): Used for simple data retrieval with explicit parameters, such as scheduled queries for daily_reports/get or hourly_reports/get reports.

Key workflow steps include:
1. Intent Judgment and Completion: Extracts supplementary information from context, asking clarifying questions only when necessary (max one round).
2. Execution and Response: Defaults to Chat mode, with script-driven execution, automatic Token detection, and handling of streaming output. Output prefixes like [RESULT] indicate results, [RESULT_FILE] indicates ultra-long content saved to files, which must be read for complete delivery.
3. Multi-round Iteration: Complex issues converge through multiple calls, e.g., first querying data then analyzing accounts, ultimately providing optimization recommendations.

Core considerations:
- Runtime requires Node.js ≥ 18; Token management is automated, with guidance for configuration when missing/expired.
- Material Link Rule: In creative inspiration and review scenarios, <img> media tags and URLs must be preserved intact without truncation or substitution.
- Ultra-long results are automatically saved via [RESULT_FILE]; reading the file is essential to ensure full content delivery.

Applicability and Considerations

• Environment Dependencies: Requires local Node.js runtime, no npm packages; Token issues are handled via exit codes 2/3 with documented steps.
• Mode Selection: API mode is limited to supported report queries, while Chat mode is broader but may require multi-round interactions for complexity (typically 3–5 rounds).
• Result Handling: Single-round results are output as-is in Markdown; multi-round syntheses can be polished but must maintain data integrity. Ultra-long results must be read from saved files to avoid link loss.
• Execution Standards: Scripts drive everything—avoid manually constructing HTTP requests; inform users of expected wait times before calls, and do not set overly short timeouts.