Browse documentation

Organize

Create crypto alerts and scheduled AI market research

Set price, indicator, drawing and screener alerts, or schedule recurring read-only AI research tasks in Basis.

9 min readUpdated 2026-08-224 sections

Use an alert for a condition and a task for an investigation

NeedUse
Price crosses a levelPrice or drawing alert
Indicator crosses a valueTechnical alert
A coin newly matches a screenScreener monitor
Daily BTC regime briefScheduled task
Check derivatives, news and whales togetherScheduled task
Notify on meaningful change from last runScheduled task with change-focused instructions

Build a precise alert

  1. 01

    Choose the source

    Select price, indicator, drawing or multi-condition logic.

  2. 02

    Choose the operator

    Crossing, crossing up/down, greater/less than and range conditions describe different events.

  3. 03

    Choose evaluation timing

    Use bar-close evaluation when intrabar noise should not trigger the rule.

  4. 04

    Name the invalidation

    An alert title should state what happened and why it matters.

  5. 05

    Enable delivery

    Browser and push notifications require explicit permission; the agent cannot grant it for you.

Schedule recurring agent research

Scheduled tasks run read-only research with Basis closed and preserve an auditable run history. They can use technicals, derivatives, news, calendar and supported whale context. They cannot place orders, mutate the chart, read private journal content or use your browser-only model key.

  1. 01

    Write the objective

    Describe the decision and data families, not just “analyze BTC.”

  2. 02

    Set cadence and timezone

    Choose a schedule that matches the market session or event.

  3. 03

    Define meaningful change

    Tell the task what difference deserves a new conclusion.

  4. 04

    Review the first run

    Confirm coverage, timestamps and caveats before relying on future runs.

Automation needs explicit failure states

  • A missed run must remain visible as missed, not as “no signal.”
  • Stale data should lower confidence and appear in the result.
  • Notification delivery is separate from task execution; inspect run history.
  • Pause schedules that no longer map to an active decision.
  • Do not create multiple overlapping tasks that repeat the same expensive research.