CrewAI Agent Examples

Knowledge-base and Product Hunt agents implemented with CrewAI and FastAPI. Both services expose NDJSON streaming APIs and surface tool calls in the stream, matching the behavior of the LangGraph examples while staying in CrewAI format.

• knowledge_agent - ingest docs, semantic search with Chroma, CrewAI-backed answers, and a streaming /stream.
• product_hunt_agent - Product Hunt launch assistant with GraphQL, Algolia search tools, and celebratory confetti payloads.

Prerequisites

• Python 3.10 or newer (3.11 recommended)
• OPENAI_API_KEY with access to GPT-4o or a compatible model
• Optional: PRODUCTHUNT_API_TOKEN for live Product Hunt data (the agent still replies with empty datasets if missing)

Setup

python3 -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install -e .

Environment variables can live in a project-level .env or service-specific .env inside knowledge_agent/ or product_hunt_agent/.

# Shared
OPENAI_API_KEY=sk-...
# OPENAI_BASE_URL=https://api.openai.com/v1
# KNOWLEDGE_OPENAI_MODEL=gpt-4o-mini
# KNOWLEDGE_EMBEDDING_MODEL=text-embedding-3-small

# Product Hunt agent
PRODUCTHUNT_API_TOKEN=phc-...
# PRODUCTHUNT_DEFAULT_TIMEZONE=America/New_York

Knowledge Agent

uvicorn knowledge_agent.main:app --host 0.0.0.0 --port 8000 --reload

Key endpoints:

• POST /api/tools/ingest - ingest text/markdown/URLs/uploads into a namespace (defaults to default).
• POST /api/tools/searchDocs - semantic search over the namespace (Chroma + OpenAI embeddings).
• POST /api/agents/knowledge/generate - non-streaming answer via a CrewAI agent.
• POST /stream - NDJSON stream with text + tool events.

Example ingestion:

curl -X POST http://localhost:8000/api/tools/ingest \
  -H "Content-Type: application/json" \
  -d '{
        "namespace": "default",
        "sources": [
          { "type": "url", "value": "https://docs.crewai.com/" },
          { "type": "markdown", "title": "Notes", "value": "# CrewAI Rocks" }
        ]
      }'

Streaming example:

curl -N http://localhost:8000/stream \
  -H "Content-Type: application/json" \
  -d '{
        "thread_id": "thread_1",
        "run_id": "run_001",
        "messages": [
          { "role": "user", "content": "Summarize the CrewAI agent lifecycle." }
        ]
      }'

Streaming format (NDJSON):

• text_start -> text_delta chunks -> text_end -> done
• Tool events: tool_call_start, tool_call_args, tool_call_end, tool_result
• Payloads include message_id, thread_id, run_id, and content for deltas.

Greeting behavior:

• If the user says “hello” (or equivalent), the agent responds with a greeting and skips tool calls.

Product Hunt Agent

uvicorn product_hunt_agent.main:app --host 0.0.0.0 --port 8001 --reload

Key endpoints:

• GET /api/top, /api/top-week, /api/top-range - Product Hunt GraphQL lookups.
• GET /api/search - Product Hunt Algolia search.
• POST /api/chat - non-streaming chat with the CrewAI agent and tool calls.
• POST /stream - NDJSON stream with text + tool events.

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 format matches the knowledge agent (text_start -> text_delta -> text_end -> done with tool call events and message/run/thread IDs).

Greeting behavior:

• If the user says “hello” (or equivalent), the agent responds with a greeting and skips tool calls.

Streaming Notes

Both /stream routes emit newline-delimited JSON (application/x-ndjson) with thread_id, run_id, type, and content. Tool call events are emitted when the CrewAI agent invokes tools, and final text is filtered to avoid leaking internal reasoning.