Building a Travel AI Agent With Agno and Qdrant (Phidata's Successor)

Description and use

Building a travel agent with Agno and Qdrant: fixing outdated imports, storing live prices via tool calls, improving retrieval with section‑based chunking and metadata, achieving 96% price accuracy and 5.2 s median response time.

Problem

Phidata tutorials from 2024 no longer install because the package became Agno in 2025 and imports changed, causing confusion for developers building travel agents.

Outcome

The revised stack delivers 100% real hotel IDs, 96% price accuracy, 83% relevant retrieval chunks, and a median response time of 5.2s, with fewer corrections needed.

Workflow

• 1. Set up Qdrant collection "travel-notes" with OpenAI embeddings.
• 2. Chunk documents by section, include heading, and add metadata fields city and doc_type.
• 3. Configure Agno Agent with OpenAIChat model gpt-5 and tools search_hotels, hotel_detail, search_notes.
• 4. Add instruction lines to enforce factual answers, single question on missing fields, and quote prices with date.
• 5. Use MCP server configuration for searchHotels, getHotelDetail, getHotelSearchTags tools.
• 6. Merge facts from vector and tool results, attaching prose only when hotel_id matches.
• 7. Measure metrics: 100% real hotel ID, 96% price accuracy, 83% relevant chunks, 0.4 corrections, 5.2s median response.

Evidence

Phidata became Agno in 2025, the imports moved, and half the travel-agent examples floating around still use the old names.

Source

The mistake I made twice: putting live prices in the vector store. Embeddings are snapshots. A price embedded last Tuesday is a wrong price with confident retrieval.

Source

Chunking is where retrieval quality is won. I split by section, not by token count, and I keep the heading in every chunk.

Source

Three instruction lines, and the third one does more work than it looks like. "Quote the date they were checked" makes stale data visible.

Source

Only attach prose from the vector store when it is keyed to the same hotel ID that came back from the tool.

Source