Review agent
The Review agent — pulling evidence from real guests
Extracts evidence from reviews that directly matches the traveller's needs, such as quiet rooms or strong Wi-Fi.
Extracts evidence from reviews that directly matches the traveller's needs, such as quiet rooms or strong Wi-Fi.
What it does
The Review agent scans hotel reviews and extracts evidence tied to the traveller's stated needs. If the user wants a quiet room, the Review agent surfaces specific review snippets that mention noise or quietness. If they want fast Wi-Fi, it pulls evidence on connection quality. The output is review_fact entries linked to the agent that requested them.
Example inputs
- "quietness for sleep"
- "Wi-Fi quality for work"
- "breakfast quality for family"
Outputs
review_fact list with: category, snippet, sentiment, source review
Try the live booking flow
Extracts evidence from reviews that directly matches the traveller's needs, such as quiet rooms or strong Wi-Fi. — much of this is already running under the hood on the current IMPT booking flow.
Search hotels →Other agents
Frequently asked
What does the Review agent actually do?
Extracts evidence from reviews that directly matches the traveller's needs, such as quiet rooms or strong Wi-Fi.
Does it run on every search?
Yes. The orchestrator dispatches all 10 agents in parallel for every search. Each agent has a time budget; if one runs slow, partial results stream into the UI while it finishes.
What if it can't find an answer?
It returns 'unknown' with confidence. The ranking engine uses the absence of evidence as a signal — a hotel where the Amenities agent can't validate Wi-Fi will rank lower than one where it can.
Where do I see this in production?
app.impt.io. The Swarm UI surfaces what each agent found — Phase 1 build in progress.