Lookouts - Scira

Pro feature - Lookouts require a Scira Pro subscription

Lookouts are scheduled research agents that automatically run queries on a recurring basis. Each run creates a new chat with full research results and sends you an email summary—perfect for monitoring topics, tracking trends, and staying informed on areas that matter to you.

What are Lookouts?

A Lookout is a scheduled research query that runs automatically based on a cron schedule. Think of it as setting up a recurring Google Alert, but powered by Scira’s full AI research capabilities with deep analysis, citations, and multi-source synthesis. Key characteristics:

Automated execution: Runs on schedule without manual intervention
Full research capability: Uses Extreme search mode by default
Email notifications: Sends results summary to your inbox
Chat history: Each run creates a new chat you can review
Run tracking: View history of all executions with metrics

Use Cases

Market & Competitor Intelligence

Title: "Weekly AI Chip Market Analysis"
Prompt: "Analyze latest developments in AI chip manufacturers, 
including NVIDIA, AMD, Intel, and emerging players. Focus on 
product announcements, market share, and analyst opinions."
Frequency: Weekly (Monday 9 AM)

Research Topic Monitoring

Title: "Daily Fusion Energy Updates"
Prompt: "Latest developments in fusion energy research, 
including breakthrough experiments, funding announcements, 
and commercial progress."
Frequency: Daily (8 AM)

Industry Briefings

Title: "Monthly SaaS Trends Report"
Prompt: "Comprehensive analysis of SaaS industry trends: 
pricing models, emerging categories, M&A activity, and 
market leaders. Include statistics and expert commentary."
Frequency: Monthly (1st of month, 6 AM)

Regulatory Tracking

Title: "EU AI Regulation Updates"
Prompt: "Track EU AI Act implementation, new regulations, 
compliance requirements, and enforcement actions. Include 
official sources and legal expert analysis."
Frequency: Weekly (Friday 5 PM)

Executive Briefings

Title: "Daily Tech News Digest"
Prompt: "Summarize top technology news: major product launches, 
M&A, funding rounds, and industry shifts. Focus on enterprise 
technology and AI developments."
Frequency: Daily (weekdays, 7 AM)

Creating a Lookout

Database Schema

Lookouts are stored in the lookout table (/lib/db/schema.ts:286):

export const lookout = pgTable('lookout', {
  id: text('id').primaryKey(),
  userId: text('user_id').notNull(),
  title: text('title').notNull(),
  prompt: text('prompt').notNull(),
  frequency: text('frequency').notNull(),  // 'once', 'daily', 'weekly', 'monthly', 'yearly'
  cronSchedule: text('cron_schedule').notNull(),
  timezone: text('timezone').notNull().default('UTC'),
  nextRunAt: timestamp('next_run_at').notNull(),
  qstashScheduleId: text('qstash_schedule_id'),
  status: text('status').notNull().default('active'),  // 'active', 'paused', 'archived', 'running'
  lastRunAt: timestamp('last_run_at'),
  lastRunChatId: text('last_run_chat_id'),
  runHistory: json('run_history').default([]),
  createdAt: timestamp('created_at').notNull().defaultNow(),
  updatedAt: timestamp('updated_at').notNull().defaultNow(),
})

Frequency Options

Once: Run one time at a specific date/time

frequency: 'once'
cronSchedule: '0 9 * * *'  // 9 AM UTC on next occurrence

Daily: Run every day at a specific time

frequency: 'daily'
cronSchedule: '0 8 * * *'  // 8 AM UTC every day

Weekly: Run on specific day(s) each week

frequency: 'weekly'
cronSchedule: '0 9 * * 1'  // 9 AM UTC every Monday
// Or multiple days: '0 9 * * 1,3,5'  // Mon, Wed, Fri

Monthly: Run on specific day of month

frequency: 'monthly'
cronSchedule: '0 6 1 * *'  // 6 AM UTC on 1st of month

Yearly: Run once per year

frequency: 'yearly'
cronSchedule: '0 0 1 1 *'  // Midnight UTC on Jan 1st

Cron Schedule Format

Lookouts use standard cron syntax with timezone support:

┌───────────── minute (0-59)
│ ┌───────────── hour (0-23)
│ │ ┌───────────── day of month (1-31)
│ │ │ ┌───────────── month (1-12)
│ │ │ │ ┌───────────── day of week (0-6) (Sunday=0)
│ │ │ │ │
│ │ │ │ │
* * * * *

Examples:

'0 9 * * *'      // 9 AM every day
'30 14 * * 1'    // 2:30 PM every Monday
'0 0 1 * *'      // Midnight on 1st of every month
'0 8 * * 1-5'    // 8 AM weekdays only
'0 */6 * * *'    // Every 6 hours

Timezone prefix:

cronSchedule: 'CRON_TZ=America/New_York 0 9 * * *'  // 9 AM Eastern
cronSchedule: 'CRON_TZ=Europe/London 0 8 * * 1'     // 8 AM London time Monday

How Lookouts Work

1. Schedule Registration

When you create a Lookout:

Record is created in lookout table
Cron schedule is registered with Upstash QStash
QStash schedule ID is stored in qstashScheduleId
Next run time is calculated and stored in nextRunAt

2. Scheduled Execution

At the scheduled time:

QStash triggers: Sends POST request to /api/lookout endpoint
Pro verification: Checks user has active Pro subscription
Status update: Sets lookout status to 'running'
Chat creation: Generates new chat with title “Scheduled: ”
Research execution: Runs Extreme search with the prompt

3. Research Process

The lookout endpoint (/app/api/lookout/route.ts) executes research using: Model: scira-grok-4-fast-think (Grok 4 with reasoning) System prompt (/app/api/lookout/route.ts:200):

3-page research paper format
Mandatory citations for every claim
Markdown formatting required
Starts with ”## Key Points” section
Deep analysis with multiple sources

Tool: extreme_search tool only

Autonomous research planning
Multi-step web and X search
Code execution if needed
Chart generation support

Step limit: Maximum 2 tool calls to prevent runaway costs

4. Result Collection

After research completes:

Message saving: User message and assistant response saved to chat
Title generation: AI generates descriptive title for the chat

Metrics tracking:

runHistory.push({
  runAt: string,          // ISO timestamp
  chatId: string,         // Link to chat
  status: 'success' | 'error' | 'timeout',
  error?: string,         // Error message if failed
  duration: number,       // Milliseconds
  tokensUsed: number,     // Total tokens
  searchesPerformed: number  // Extreme search count
})

Extreme search usage: Incremented in user’s usage tracking

5. Email Notification

Email is sent using Resend (/lib/email.ts): To: User’s registered email Subject: “Your Scira Lookout has completed: ” Content:

Lookout title
Research summary (first 2000 chars of response)
Link to full chat results
Run timestamp

Template: React Email components for rich formatting

6. Next Run Calculation

For recurring lookouts (frequency !== 'once'):

Parse cron schedule with cron-parser
Calculate next occurrence from current time
Update nextRunAt field
Status returns to 'active'

For one-time lookouts:

Status set to 'paused' after completion
Not automatically deleted (can be manually re-run)

Run History

Each lookout maintains a complete run history stored as JSON:

type RunHistoryEntry = {
  runAt: string;              // ISO date string
  chatId: string;             // Link to generated chat
  status: 'success' | 'error' | 'timeout';
  error?: string;             // Error message if failed
  duration?: number;          // Execution time in milliseconds
  tokensUsed?: number;        // Total tokens consumed
  searchesPerformed?: number; // Number of extreme searches
}

Example run history:

[
  {
    "runAt": "2026-03-04T14:00:00.000Z",
    "chatId": "01932e7d-8b5f-7890-1234-567890abcdef",
    "status": "success",
    "duration": 45230,
    "tokensUsed": 12450,
    "searchesPerformed": 1
  },
  {
    "runAt": "2026-03-03T14:00:00.000Z",
    "chatId": "01932a1b-2c3d-4e5f-6789-0abcdef12345",
    "status": "success",
    "duration": 38920,
    "tokensUsed": 10230,
    "searchesPerformed": 1
  }
]

Status States

Lookouts can be in one of four states: active

Scheduled and will run at next occurrence
Waiting for QStash trigger
Can be paused or archived

paused

Temporarily disabled
Schedule still exists in QStash but won’t execute
Can be resumed to active
One-time lookouts auto-pause after running

archived

Soft-deleted
No longer shown in active list
Schedule removed from QStash
Can be permanently deleted

running

Currently executing research
Temporary state during execution
Returns to active after completion
Prevents duplicate concurrent runs

Error Handling

Execution Errors

If research fails (/app/api/lookout/route.ts:544):

Error captured in runHistory with status 'error'
Duration recorded
Status returns to 'active' (will retry next scheduled time)
No email sent on error

Timeout Handling

Long-running research (>5 minutes):

QStash has built-in timeout
Status recorded as 'timeout' in history
Lookout returns to active for next run

Subscription Lapses

If user’s Pro subscription expires:

Lookout remains in database
Execution fails Pro check (/app/api/lookout/route.ts:133)
Returns 403 Forbidden
Schedule remains in QStash (will retry)
User must renew to resume execution

Best Practices

Prompt Design

Be specific about scope:

Good: "Analyze top 5 AI image generators by user base, 
feature set, and pricing. Include Reddit sentiment."

Bad: "What's happening with AI art?"

Include date context:

Good: "Latest developments this week in renewable energy"
Bad: "Renewable energy news"  // Will search without time context

Request specific formats:

Good: "Create comparison table of cloud providers with 
pricing, features, and region availability"

Bad: "Compare cloud providers"  // Less structured

Specify sources when relevant:

Good: "Track EU AI Act updates from official EU sources, 
legal experts, and industry commentary"

Bad: "AI regulation news"  // May miss authoritative sources

Scheduling Strategy

Match frequency to update cadence:

Breaking news topics: Daily or multiple times per day
Research areas: Weekly or bi-weekly
Market reports: Weekly or monthly
Regulatory tracking: Weekly or monthly

Consider timezone:

// Get briefing before work starts (8 AM local time)
timezone: 'America/New_York'
cronSchedule: 'CRON_TZ=America/New_York 0 8 * * 1-5'

// End of day summary (6 PM local time)
timezone: 'Europe/London'
cronSchedule: 'CRON_TZ=Europe/London 0 18 * * *'

Avoid overlapping schedules:

Stagger multiple lookouts by at least 15 minutes
Prevents concurrent extreme searches hitting limits

Cost Management

Lookouts use Extreme search by default, which:

Consumes more tokens than regular chat
Counts against Extreme search limit (Pro plan)
May incur API costs for search providers

Optimize costs:

Use appropriate frequency (don’t over-schedule)
Make prompts focused to avoid unnecessary searches
Monitor run history for token usage trends
Pause lookouts during periods you don’t need updates

QStash Integration

Lookouts use Upstash QStash for reliable scheduling:

// From package.json:
"@upstash/qstash": "^2.8.4"

Features:

Serverless cron scheduling
Automatic retries on failure
Timezone support
Delivery guarantees

Schedule management:

// Create schedule (pseudocode)
const scheduleId = await qstash.schedule({
  url: `${baseUrl}/api/lookout`,
  cron: cronSchedule,
  body: JSON.stringify({ lookoutId, prompt, userId })
})

// Delete schedule
await qstash.schedules.delete(scheduleId)

Email Integration

Emails are sent using Resend API:

// From package.json:
"resend": "^6.6.0"
"@react-email/components": "^1.0.1"

Email template (/lib/email.ts):

React Email components for rendering
Responsive HTML design
Plain text fallback
Direct link to chat results
Truncated summary (2000 chars max)

Technical Architecture

Request Flow:

QStash (scheduled)
  ↓
POST /api/lookout
  ↓
1. Verify Pro subscription
2. Set status to 'running'
3. Create new chat
4. Execute streamText with extreme_search
5. Save messages and metrics
6. Send email notification
7. Calculate next run
8. Return to 'active' status

Dependencies:

// Key dependencies for lookout functionality:
import { streamText, stepCountIs } from 'ai'
import { extremeSearchTool } from '@/lib/tools'
import { sendLookoutCompletionEmail } from '@/lib/email'
import { CronExpressionParser } from 'cron-parser'
import { updateLookoutLastRun, updateLookout } from '@/lib/db/queries'

Model configuration:

model: scira.languageModel('scira-grok-4-fast-think')
stopWhen: stepCountIs(2)  // Max 2 tool calls
activeTools: ['extreme_search']
toolChoice: 'auto'
maxRetries: 10

Limitations

Pro requirement: Lookouts are only available to Pro subscribers Execution limits:

Maximum 2 tool calls per run (prevents runaway costs)
Extreme search usage counted against Pro limits
QStash timeout after extended execution

Frequency constraints:

Minimum practical interval: ~15 minutes (QStash limitation)
Maximum history size: No hard limit but stored as JSON

Email delivery:

Requires valid email on user account
Subject to Resend rate limits
No email sent on execution errors

Next Steps

Extreme Search

Learn about the Extreme search mode used by Lookouts

Tools

Deep dive into the extreme_search tool

Get Started

Features

Guides

Integrations

​What are Lookouts?

​Use Cases

​Market & Competitor Intelligence

​Research Topic Monitoring

​Industry Briefings

​Regulatory Tracking

​Executive Briefings

​Creating a Lookout

​Database Schema

​Frequency Options

​Cron Schedule Format

​How Lookouts Work

​1. Schedule Registration

​2. Scheduled Execution

​3. Research Process

​4. Result Collection

​5. Email Notification

​6. Next Run Calculation

​Run History

​Status States

​Error Handling

​Execution Errors

​Timeout Handling

​Subscription Lapses

​Best Practices

​Prompt Design

​Scheduling Strategy

​Cost Management

​QStash Integration

​Email Integration

​Technical Architecture

​Limitations

​Next Steps

Extreme Search

Tools

Build docs developers (and LLMs) love

What are Lookouts?

Use Cases

Market & Competitor Intelligence

Research Topic Monitoring

Industry Briefings

Regulatory Tracking

Executive Briefings

Creating a Lookout

Database Schema

Frequency Options

Cron Schedule Format

How Lookouts Work

1. Schedule Registration

2. Scheduled Execution

3. Research Process

4. Result Collection

5. Email Notification

6. Next Run Calculation

Run History

Status States

Error Handling

Execution Errors

Timeout Handling

Subscription Lapses

Best Practices

Prompt Design

Scheduling Strategy

Cost Management

QStash Integration

Email Integration

Technical Architecture

Limitations

Next Steps