Model testing

Ask the DivvyAI agent.

Questions go to the real DivvyAI agent. It picks from three read-only tools, queries recorded Divvy station observations in BigQuery and explains what it finds. Every answer shows the tool calls and data behind it.

DivvyAI agent
not checked yet
Examples

Ask about recent Divvy observations

Pick an example above to fill the question box, or write your own below. Nothing is sent until you press Ask the agent.

Ready · each question is answered on its own POST /api/agent

find_capture_gapsread-only

Gaps between observed captures from yesterday’s UTC midnight to now. Both feeds can show the same delay, and no gaps doesn’t prove collection is healthy.

min_minutes = 30, limit = 20

rank_stationsread-only

Stations ranked by how often they were sampled with no bikes or no docks, or by lowest averages, for one Chicago hour across up to seven recent completed days.

metric, hour, limit, start_date, end_date, day_type, area

get_station_availabilityread-only

Hourly average bikes and docks with sample counts for an exact station ID, within the last seven UTC days. Hours are UTC.

station_id, start, end

Not supported
  • Forecasts or predictions of any kind
  • Looking up stations by name
  • History older than seven days
  • Whether a station is operating
  • Follow-ups: each question is answered on its own

Good to know

  • Each question gets a small, fixed budget of model and tool calls. If it runs out, the answer is marked incomplete.
  • Missing hours and empty averages are unknown, not zero. downtown_demo is an illustrative box, not an official boundary.
  • Questions are sent to the DivvyAI agent API and to OpenAI. The conversation stays in this tab and clears on reload.