AI Trends Scheduled Summary

A scheduled AI news aggregation agent built with the OpenAI Agents SDK on EdgeOne Makers — automatically collects, curates, scores, and generates daily trend reports from multiple sources.

Framework: OpenAI Agents SDK · Category: Scheduled · Language: TypeScript

Overview

AI Trends Scheduled Summary runs a 4-agent pipeline on a daily cron schedule (or manual trigger) to collect AI industry news from Hacker News, Dev.to, and web sources, then produces a curated Markdown trend report. The entire pipeline streams progress via SSE so users can watch items being fetched, filtered, scored, and written in real time.

• Multi-source collection — pulls from Hacker News, Dev.to, and configurable web sources (via sandbox browser scraping)
• 4-agent pipeline — Curator (filter) + Summarizer (summarize) run in parallel, followed by Analyst (score + classify) and Writer (Markdown report)
• Real-time SSE streaming — progressive content disclosure from fetch through final report, with token-level Writer streaming
• Cross-run deduplication — fingerprint-based item library tracks seenCount, firstSeenAt, lastSeenAt across scheduled runs
• Comprehensive scoring — Analyst assigns 0–100 scores based on source engagement, content quality, and AI-relevance

Environment Variables

Variable	Required	Description
AI_GATEWAY_API_KEY	Yes	Model gateway API key. Use your Makers Models API Key, or any OpenAI-compatible provider key.
AI_GATEWAY_BASE_URL	Yes	Gateway base URL. For Makers Models, use https://ai-gateway.edgeone.link/v1.
AI_GATEWAY_MODEL	No	Model ID. Defaults to @makers/minimax-m2.7.

This template follows the OpenAI-compatible standard — you can point these variables at Makers Models or any other compatible gateway / provider.

How to get AI_GATEWAY_API_KEY

1. Open the Makers Console.
2. Sign in and enable Makers.
3. Go to Makers → Models → API Key and create a key.
4. Copy it into AI_GATEWAY_API_KEY (set AI_GATEWAY_BASE_URL to https://ai-gateway.edgeone.link/v1).

Built-in models (@makers/deepseek-v4-flash, @makers/hy3-preview, @makers/minimax-m2.7) are free and rate-limited — great for prototyping. For production, bind your own provider key (BYOK) in the console.

Provider fallbacks

The code reads environment variables with the following priority chain:

LLM_API_KEY → AI_GATEWAY_API_KEY → OPENAI_API_KEY
LLM_BASE_URL → AI_GATEWAY_BASE_URL → OPENAI_BASE_URL
LLM_MODEL → AI_GATEWAY_MODEL → @makers/minimax-m2.7

You can set any of these depending on your provider preference.

Local Development

Prerequisites: Node.js ≥ 18, npm

npm install
cp .env.example .env
edgeone makers dev

Open http://localhost:8080/agent-metrics for the local observability panel.

Project Structure

ai-trends-scheduled-summary/
├── agents/
│   └── ai-trends/
│       ├── run.ts              # /ai-trends/run — main pipeline entry (SSE stream)
│       ├── stop.ts             # /ai-trends/stop — abort a running pipeline
│       ├── _model.ts           # 4-agent definitions, prompts, streaming logic
│       ├── _sources.ts         # Data collection (HN, Dev.to, sandbox browser)
│       ├── _items.ts           # Item library: fingerprinting, merge, dedup
│       ├── _memory.ts          # Platform store persistence (reports + items)
│       ├── _storage.ts         # File-system fallback persistence
│       ├── _report.ts          # Report assembly helpers
│       ├── _http.ts            # Request/response utilities
│       └── _types.ts           # Shared Zod schemas & TypeScript types
├── cloud-functions/
│   └── ai-trends/
│       ├── latest/index.ts     # GET /ai-trends/latest
│       ├── history/index.ts    # GET /ai-trends/history
│       ├── detail/index.ts     # POST /ai-trends/detail
│       ├── delete/index.ts     # POST /ai-trends/delete
│       └── health/index.ts     # GET /ai-trends/health
├── src/                        # Frontend (React + Vite)
│   ├── App.tsx                 # Main UI: LiveFeed, PipelineBar, ReportDrawer
│   ├── api.ts                  # SSE client & REST helpers
│   ├── i18n.tsx                # Chinese/English i18n
│   ├── MarkdownReport.tsx      # Markdown renderer
│   ├── reportModel.ts          # Frontend report normalization
│   └── types.ts                # Frontend type definitions
├── edgeone.json                # Agent runtime & schedule configuration
└── package.json

Files prefixed with _ are private modules — not exposed as public routes by EdgeOne.

How It Works

The agent runs as a session-mode runtime under agents/. Requests sharing the same conversation_id are routed to the same instance (and sandbox, when available).

Pipeline Flow

1. Trigger — either via cron schedule (0 9 * * * daily) or manual POST to /ai-trends/run. The request includes sources (default: hackernews, devto, web) and limit.

2. Fetch & Merge — collects candidates from configured sources. Hacker News and Dev.to use public APIs; web sources use the sandbox browser (context.sandbox.browser.goto + evaluate) for JS-rendered pages. Candidates are deduplicated against the item library using URL/title fingerprints.

3. Curator + Summarizer (parallel) — two agents run concurrently via Promise.allSettled:

   • Curator filters irrelevant items, assigns categories (AI Agent, LLM, Multimodal, etc.), and decides keep/drop.
   • Summarizer generates 1–2 sentence Chinese summaries for each item.

4. Analyst — scores each item 0–100 (weighted: 40% quality, 30% heat, 30% relevance), groups by category, identifies new/active/single status, and optionally deep-dives into 2–3 top articles via fetch_url sandbox tool.

5. Writer (token-streaming) — generates a structured Markdown report streamed token-by-token to the client. Filters <think> tags in real time. Falls back to non-streaming retry on connection failure.

6. Persist — the final report is saved to context.store (platform memory). If unavailable, falls back to file-system storage.

SSE Streaming Protocol

The /ai-trends/run endpoint returns an SSE stream with typed events:

Event type	Purpose
stage	Pipeline stage status transitions (running / done / failed)
items	Progressive content snapshots (fetched → curated → summarized)
analysis	Analyst output (categories, scores, keyInsight)
progress	Keepalive during long LLM calls (emitted every 8s)
token	Writer Markdown tokens for live-typing UX
complete	Terminal event with the full TrendReport payload

Key Design Decisions

• Graceful degradation — if Writer fails, report is assembled from Analyst output; if Analyst fails, a code-generated fallback is used.
• AbortSignal — threaded through all stages; users can stop generation mid-pipeline.
• Conversation-scoped storage — reports and item library are stored per conversation via context.store.appendMessage / getMessages.

Runtime Configuration

From edgeone.json:

• agents.timeout: 1200s (20 min max pipeline duration)
• agent.sandbox.timeout: 300s (sandbox lifetime for browser scraping)
• schedules[0].cron: 0 9 * * * (daily at 01:00 UTC)

Route Summary

Route	Method	Description
/ai-trends/run	POST	Start pipeline (SSE stream)
/ai-trends/stop	POST	Abort running pipeline
/ai-trends/latest	GET	Latest report
/ai-trends/history	GET	Report history list
/ai-trends/detail	POST	Specific report by runId
/ai-trends/delete	POST	Delete report by runId

The conversation_id is passed via the makers-conversation-id request header.

Resources

• Makers Agents Documentation
• Quick Start: Agent Development
• Makers Models

License

MIT