Luma / A WORKFLOW WITH YOUR AGENT

Find the events worth making time for.

You follow many event calendars. Finding something relevant means opening each one, checking dates and locations, then comparing registration details across tabs.

ASK FOR THE OUTCOME

Give your agent a clear starting point.

“Read the selected Luma calendars I follow in your signed-in browser. Capture their accessible upcoming events in PocketContext. Shortlist relevant events in my city for the next seven days, with times, source links and availability as observed. Flag incomplete listings.”

An example request for an agent configured to use PocketContext. Reading this page does not run it.

CAPTURE ONCE, THEN REUSE

From browsing to stored context.

Read the selected sources

Sign in to Luma in the agent’s browser and choose which followed calendars matter. The agent reads their descriptions and accessible upcoming events, retaining source links and the time of the check.

Give the information structure

An entity page per calendar holds its description and captured events. Properties such as city, followed and captured_at let your agent filter the collection. Events can have separate linked pages when you need to query them individually.

ILLUSTRATIVE PAGEentity

City Builders · followed calendar

source_app: luma
city: Berlin
followed: true
captured_at: 2026-11-03

See the Markdown body

## About this calendar
Community description and source link.

## Upcoming events
Captured listings, dates and availability, with evidence.

Properties for querying. Markdown for the detail. Citations and links connect the evidence.

AN EXAMPLE OF THE RESULT

A shortlist for your week

A useful brief makes the evidence and the uncertainty visible. The following is fictional reader-style content, not a live account or a completed capture.

Build

A local product workshop matches your interest in practical AI projects. Registration was open at capture time.

Read example evidence 1

Fictional event capture: “Product Workshop — Berlin. Bring a small AI project. Registration open.” Captured November 3, 2026.

Meet

A founders’ breakfast offers a chance to meet other early-stage teams. Attendance requires approval.

Read example evidence 2

Fictional event capture: “Founders Breakfast — Berlin. Meet other early-stage teams. Apply to attend; organizer approval required.”

Check first

One calendar’s event list was incomplete. Refresh it and verify live availability before registering.

Read example evidence 3

Fictional capture note: additional events were indicated, but the full list was not accessible during this capture.

THE NEXT QUESTION

Query the knowledge.
Skip the repeated scrolling.

Ask your agent to list followed calendars in a city, find captured events on a topic or identify calendars due for another check. Query event dates directly when events have their own structured pages.

SQL reads, full-text search and authenticated REST APIs let the agent retrieve stored context directly. The browser reader gives you published pages, properties, citations, originals and revision history.

Keep the browser for changes

Computer use can be slow and token-intensive. Reusing captured information avoids repeating the same navigation for each question. Capture and refreshes still cost time and model usage; savings are workflow-dependent.

Event times, venues and availability can change after capture. Use the shortlist to decide what to inspect, then check the live event page before registering or travelling.

AFTER A SUCCESSFUL FIRST CAPTURE

Make the useful part repeatable.

“Every evening, revisit these selected Luma calendars. Capture new or changed event details, update their pages and prepare a shortlist for the next seven days. Include when availability was checked and any missing listings. Do not register for events.”

Use a scheduler supplied by your agent, such as Grok Bot routines . Start with a small selection, verify the results and choose a frequency that fits how quickly the information changes.

The agent supplies browser access and scheduling; PocketContext has no built-in connector for this service or recurring scheduler. Sessions can expire and captures can be incomplete. This routine has not been rehearsed end to end.

KEEP THE CONTEXT USEFUL

Readable detail, queryable properties.

These sample properties are conventions your agent can adopt, not dedicated service schemas. Record source URLs and capture dates alongside them. Markdown bodies, properties, citations and history live together in the database. The agent must verify identity matches and create links between related pages.

Capture only the material you choose. External model providers receive the context your agent sends. A citation helps you inspect a claim; it does not prove the conclusion is correct.

Explore the shared data model

Start with a small collection.

Bring one source, ask for a useful brief, and inspect the evidence.

Set up PocketContext Explore more workflows

Fictional example · PocketContext 0.24.0 · Product data model