AI Day Planner with Parallel APIs and Gemini in Kestra

id: ai-plan-my-day
namespace: company.team
description: |
  Geocode a city, fetch weather and air quality in parallel, and ask an
  AI agent to write a short daily plan, with retries, an SLA, and a
  Slack alert on failure.

inputs:
  - id: city
    type: STRING
    displayName: City name
    defaults: Austin

tasks:
  - id: log_input
    type: io.kestra.plugin.core.log.Log
    description: Record which city this run is planning for.
    message: "Incoming city: {{ inputs.city }}"

  - id: geocode_city
    type: io.kestra.plugin.core.http.Request
    description: Resolve the city name to coordinates with Open-Meteo geocoding.
    uri: "https://geocoding-api.open-meteo.com/v1/search?name={{ inputs.city
      }}&count=1&language=en&format=json"
    method: GET

  - id: parse_geo
    type: io.kestra.plugin.scripts.python.Script
    description: Extract latitude and longitude from the geocoding response.
    dependencies:
      - kestra
    script: |
      import json
      from kestra import Kestra

      data = json.loads("""{{ outputs.geocode_city.body }}""")
      if not data.get("results"):
          raise RuntimeError("Invalid response from Open-Meteo Geocoding.")
      lat = data["results"][0]["latitude"]
      lon = data["results"][0]["longitude"]
      print(json.dumps({"lat": lat, "lon": lon}))
      Kestra.outputs({
        "lat": lat,
        "lon": lon,
      })

  - id: parallel_tasks
    type: io.kestra.plugin.core.flow.Parallel
    description: Fetch weather and air quality concurrently.
    allowFailure: true
    allowWarning: true
    tasks:
      - id: weather_api
        type: io.kestra.plugin.core.http.Request
        uri: "https://api.open-meteo.com/v1/forecast?latitude={{
          outputs.parse_geo.vars.lat | string }}&longitude={{
          outputs.parse_geo.vars.lon | string
          }}&current=temperature_2m,wind_speed_10m"
        method: GET

      - id: air_quality_api
        type: io.kestra.plugin.core.http.Request
        uri: "https://air-quality-api.open-meteo.com/v1/air-quality?latitude={{
          outputs.parse_geo.vars.lat | string }}&longitude={{
          outputs.parse_geo.vars.lon | string
          }}&current=pm2_5,pm10,us_aqi,ozone"
        method: GET

  - id: day_brief
    type: io.kestra.plugin.ai.agent.AIAgent
    description: Turn the raw API data into a short, friendly plan for the day.
    systemMessage: |
      You are a concise assistant. Use the provided JSON to write a short, friendly plan.
      Keep it to 3 to 5 sentences. Include any cautions from the data.
    prompt: |
      City: {{ inputs.city }}
      weather: {{ outputs.weather_api.body ?? "not available!" }}
      air quality: {{ outputs.air_quality_api.body ?? "not available!" }}
      Write a brief suggestion for what someone should do today.

  - id: log_output
    type: io.kestra.plugin.core.log.Log
    description: Keep the plan in the execution logs.
    message: "{{ outputs.day_brief.textOutput }}"

triggers:
  - id: incoming_webhook
    type: io.kestra.plugin.core.trigger.Webhook
    description: Any HTTP caller can request a day plan on demand.
    key: "{{ secret('PLAN_MY_DAY_WEBHOOK_KEY') }}"

errors:
  - id: slack_fail
    type: io.kestra.plugin.slack.notifications.SlackIncomingWebhook
    description: Alert the channel if any task fails.
    url: "{{ secret('SLACK_WEBHOOK') }}"
    payload: |
      {
        "text": ":rotating_light: Flow {{ flow.id }} failed on {{ inputs.city }} - Error: {{ error.message }}"
      }

sla:
  - id: must_finish_fast
    type: MAX_DURATION
    duration: PT10M
    behavior: CANCEL
    labels:
      sla: miss
      reason: "Exceeded allowed execution time"

outputs:
  - id: brief
    type: STRING
    value: "{{ outputs.day_brief.textOutput }}"

pluginDefaults:
  - type: io.kestra.plugin.ai.agent.AIAgent
    values:
      provider:
        type: io.kestra.plugin.ai.provider.GoogleGemini
        modelName: gemini-3.5-flash-lite
        apiKey: "{{ secret('GEMINI_API_KEY') }}"
      allowFailure: true
      configuration:
        chatConfiguration:
          logRequests: true
      retry:
        type: constant
        interval: PT2S
        maxAttempts: 2

  - type: io.kestra.plugin.core.http.Request
    values:
      headers:
        Accept: "application/json"
      allowFailure: true
      retry:
        type: constant
        interval: PT3S
        maxAttempts: 3

A compact, end-to-end pattern for AI-assisted API orchestration. This Kestra blueprint takes a city name, resolves it to coordinates with the Open-Meteo geocoding API, fans out to the weather and air quality APIs in parallel, and hands the raw JSON to an AI agent that writes a 3 to 5 sentence plan for the day, including cautions such as high PM2.5 or strong wind. It demonstrates the operational scaffolding real AI pipelines need: per-task retries, graceful degradation when one API fails, an execution SLA, a failure alert to Slack, and a typed flow output that downstream flows or callers can consume.

How it works

The incoming_webhook trigger (io.kestra.plugin.core.trigger.Webhook) lets any HTTP caller request a plan on demand, secured by PLAN_MY_DAY_WEBHOOK_KEY. The flow also runs manually with the city input, defaulting to Austin.
The geocode_city task (io.kestra.plugin.core.http.Request) resolves the city name, and parse_geo (io.kestra.plugin.scripts.python.Script) extracts latitude and longitude, failing fast with a clear error when the city is unknown.
The parallel_tasks task (io.kestra.plugin.core.flow.Parallel) calls weather_api and air_quality_api concurrently. Both requests inherit retry and header policy from pluginDefaults and are allowed to fail individually.
The day_brief task (io.kestra.plugin.ai.agent.AIAgent) prompts gemini-3.5-flash-lite with both response bodies, substituting "not available!" for any missing data so the agent still produces a useful plan on partial failures.
log_output records the answer, the flow exposes it as the brief output, the errors branch posts to Slack on failure, and a MAX_DURATION SLA cancels runs over 10 minutes.

What you get

A working webhook-to-AI pipeline in about 100 lines of declarative YAML.
Parallel API fan-out with retries and partial-failure tolerance baked in.
A reusable pattern: geocode, enrich in parallel, summarize with an agent, alert on failure.
A typed brief flow output that subflows, apps, or API callers can consume directly.

Who it's for

Teams learning how to combine HTTP orchestration with the Kestra AI agent.
Developers building personal assistant or notification products on public APIs.
Solutions engineers who need a small but production-shaped AI demo.

Why orchestrate this with Kestra

The interesting part of this flow is not any single API call, it is the failure handling between them. Kestra provides Parallel for concurrent enrichment, pluginDefaults to declare retry and header policy once for every HTTP request, allowFailure for graceful degradation, an errors branch for alerting, and SLA enforcement, all without a line of orchestration code. The AIAgent task turns the LLM call into an observable task with logged requests, so you can see exactly what the model was given when a plan looks off.

Prerequisites

A Google AI Studio API key with access to Gemini models.
A Slack incoming webhook URL for failure alerts.
No API keys are needed for the Open-Meteo geocoding, weather, and air quality endpoints.

Secrets

GEMINI_API_KEY: Google Gemini API key used by the AIAgent task.
SLACK_WEBHOOK: Slack incoming webhook URL for the failure alert.
PLAN_MY_DAY_WEBHOOK_KEY: shared secret guarding the Webhook trigger.

Quick start

Add the secrets above to your Kestra namespace.
Run the flow manually with the default city and read the plan in the logs.
POST to the webhook URL with the configured key to trigger it from any client.
Change the city default or pass it in the webhook payload to plan for other locations.

How to extend

Add a Schedule trigger to deliver the plan every morning instead of on demand.
Post the brief to Slack, email, or Telegram with the corresponding notification task.
Add more parallel sources such as calendar events or commute times before the agent call.
Return structured JSON from the agent with responseFormat and branch on fields like a rain flag.
Cache geocoding results in the KV store to skip repeat lookups for the same city.

AI Day Planner from Weather and Air Quality APIs

How it works

What you get

Who it's for

Why orchestrate this with Kestra

Prerequisites

Secrets

Quick start

How to extend

Links

Tasks

Orchestrate with Kestra

Share this Blueprint

Create a Blueprint

New to Kestra?

AI Day Planner from Weather and Air Quality APIs

How it works

What you get

Who it's for

Why orchestrate this with Kestra

Prerequisites

Secrets

Quick start

How to extend

Links

Tasks

Orchestrate with Kestra

Share this Blueprint

Create a Blueprint

More AI Blueprints

Convert Natural Language Prompts to Structured Todoist Tasks with Deepseek AI

Generate a book chapter summary using Perplexity AI and email it.

AI Document Data Extraction: Parse PDFs and Web Pages into Structured JSON with OpenAI

AI Annual Report Finder and Data Strategy Summary

Generate and Send a Personalized AI-Powered Daily Digest Email Every Morning

Summarize Text with an AI Agent, Generate Multilingual Summaries in Short, Medium, or Long Format

New to Kestra?