Atlas AI — Brand Launch Copilot

面向中国消费品牌出海的 AI 品牌策划智能体，属于 SynNovator「Q3W3 第三轮｜品牌出海场景赛道」的 AI 品牌策划子赛道。

Atlas 把一次品牌 Brief 编排成可审核、可交付的海外市场启动包：市场扫描、受众地图、品牌定位、本地化检查、内容资产和 90 天启动计划。项目重点展示 Wave 3 要求的完整 Agent 运行入口、执行轨迹、结构化输出和人工审核节点。

What it does

• 接收品牌、产品、目标市场、首发渠道和增长目标
• 通过 6 个 specialist skills 编排一次完整分析：Market scan、Audience map、Positioning、Localization check、Content kit、Launch plan
• 返回 Launch readiness、品牌定位、目标受众、市场信号和 Evidence
• 生成 Brand positioning、Product page、Creator kit、90-day playbook 四类可审核交付物
• 支持真实 OpenAI-compatible API；模型输出异常时自动修复 JSON，接口不可用时保留 Demo fallback
• 页面提供 Brand strategy、Content studio、Market signals、Launch cockpit 四个工作台

Run locally

Requirements: Python 3.9+ and an OpenAI-compatible Chat Completions endpoint.

cd atlas-ai
python -m pip install -r requirements.txt
$env:OPENAI_API_KEY="your-api-key"
$env:OPENAI_BASE_URL="https://api.oaipro.com/v1"
$env:OPENAI_MODEL="gpt-5.5"
$env:PORT="49173"
python server.py

Open http://localhost:49173. The server also auto-detects a local D:\xds\2 key during the local demo only; use OPENAI_API_KEY in any shared or deployed environment.

Demo brief

Brand: NOVA Objects
Product: Modular desk light
Market: Japan
Channel: DTC + TikTok Shop
Objective: Build a premium, design-led work-from-home brand for creators

Click Run launch intelligence. A successful live response is marked Live API; the output updates the score, insights, execution trace, evidence and deliverable cards. Clicking a deliverable opens its generated content for review and approval.

Architecture

Browser brief form
        |
        v
POST /api/launch (server.py)
        |
        +--> OpenAI-compatible Chat Completions API
        |       |
        |       +--> strict Atlas JSON schema
        |       +--> JSON validation / repair request
        |
        +--> fallback sample when no key or upstream failure
        |
        v
Launch cockpit + trace + evidence + reviewable assets

The browser never receives the API key. server.py owns the provider call and validates the required fields before returning a response.

Wave 3 updates

Compared with a concept-only workflow, this submission adds a runnable Agent entry point, six-step specialist trace, structured JSON validation, automatic repair, evidence and assumption display, reviewable deliverables, human approval, live API mode and a resilient demo fallback.

Submission evidence

See docs/demo-script.md for the recommended three-minute walkthrough and docs/architecture.md for the implementation details. The live Demo URL and the Pull Request URL are entered on the SynNovator activity page after creating a Fork.

Repository rules

SUBMISSIONS.md and submissions.json are maintained automatically by the official track workflow. Do not edit them by hand.