Skip to main content
Refreshed for the latest ai-agent-crew-ai-examples codebase: run the product_hunt_agent FastAPI service, wire it to CometChat, and stream SSE updates (with optional confetti actions).

​
What you’ll build

  • A CrewAI agent with tools to get top posts, search, timeframes, and trigger confetti.
  • A FastAPI /stream endpoint emitting newline-delimited JSON (text_start, text_delta, text_end, done).
  • CometChat AI Agent wiring that consumes those SSE chunks; your UI listens for the confetti payload.
  • Streaming events follow text_start → text_delta chunks → text_end → done (errors emit type: "error").

​
Prerequisites

  • Python 3.10+ with pip
  • OPENAI_API_KEY (optionally OPENAI_BASE_URL, PRODUCT_OPENAI_MODEL)
  • Optional: PRODUCTHUNT_API_TOKEN for live GraphQL data (empty lists when missing)
  • CometChat app + AI Agent entry

​
Run the updated sample

1

Install & start

In ai-agent-crew-ai-examples/: 
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
uvicorn product_hunt_agent.main:app —host 0.0.0.0 —port 8001 —reload
2

Set env

Required: OPENAI_API_KEY. Optional: PRODUCTHUNT_API_TOKEN (GraphQL), PRODUCTHUNT_DEFAULT_TIMEZONE (default America/New_York).

​
API surface (FastAPI)

  • GET /api/top — top posts by votes (limit 1–10).
  • GET /api/top-week — rolling window (default 7 days) with limit and days.
  • GET /api/top-range — timeframe queries (timeframe, tz, limit); supports "today", "yesterday", "last_week", "last_month", or ISO dates.
  • GET /api/search — Algolia search (q, limit).
  • POST /api/chat — non-streaming CrewAI answer.
  • POST /stream — SSE stream (text_start, text_delta, text_end, done) ready for CometChat.
  • POST /api/chat payload: {"message": "…", "messages": [{ "role": "user", "content": "…" }]} (array is required if message is omitted).

​
Streaming example

curl -N http://localhost:8001/stream \
  -H "Content-Type: application/json" \
  -d '{
        "messages": [
          { "role": "user", "content": "What were the top launches last week?" }
        ]
      }'
Streaming payload shape: 
{"type":"text_start","message_id":"...","thread_id":"...","run_id":"..."}
{"type":"text_delta","content":"...","message_id":"...","thread_id":"...","run_id":"..."}
{"type":"text_end","message_id":"...","thread_id":"...","run_id":"..."}
{"type":"done","thread_id":"...","run_id":"..."}
# errors (when thrown) look like:
{"type":"error","message":"...","message_id":"...","thread_id":"...","run_id":"..."}

​
Crew internals (for reference)

Key tools in product_hunt_agent/agent_builder.py: 
@tool("getTopProducts")          # votes-ranked, clamps limit 1-10
@tool("getTopProductsThisWeek")  # rolling-week window, clamps days 1-31 and limit 1-10
@tool("getTopProductsByTimeframe")  # "today", "yesterday", "last_week", ISO, ranges; clamps limit 1-10
@tool("searchProducts")          # Algolia search (no token needed)
@tool("triggerConfetti")         # returns payload: colors, particleCount, spread, startVelocity, origin, ticks, disableSound
All tools run server-side; if PRODUCTHUNT_API_TOKEN is missing, top/timeframe queries return empty arrays but still respond cleanly (search still works via Algolia defaults).

​
Wire it to CometChat

  • Dashboard → AI Agent → BYO Agents and then Get Started / Integrate → Choose CrewAI. → Agent ID (e.g., product_hunt) → Deployment URL = your public /stream.
  • Listen for text_start/text_delta/text_end to render streaming text; stop on done.
  • When triggerConfetti returns, map the payload to your UI handler (Widget/React UI Kit). Keep API tokens server-side.