BURKI IDLE REMINDER WORKSHEET
Free planning resource | prepared October 20, 2026
Original guide: https://burki.dev/blog/burki-idle-reminder-setup

This is an unperformed review plan with fictional settings. Reading/download is free. Voice tests and production use can incur platform, provider and telephone costs. No configuration was saved and no call was run for this resource.

1. PURPOSE AND BASELINE
Business/task: _______________________________________________
Example: Alder Parts Desk; leave time to find a job reference.
Owning workspace/assistant: __________________________________
Conversation engine: ________________________________________
Must be Standard pipeline for these idle reminders. GPT Live requires idle timeout zero; it does not apply the standard idle reminder feature.
Recognition/model/voice/language: ____________________________
Intended route (browser OR exact telephone route): ___________
Saved/published version and timestamp: _______________________
Legacy live settings warning present? ________________________
Credential/funding/readiness reviewed by/date: _______________
Approved observation/test authority and budget: ______________

2. THREE DIFFERENT CONTROLS
Turn completion: whether the caller finished a thought.
Idle reminder: an eligible inactive conversation waited long enough.
Admitted duration: total session limit, independently constrained by admission/funding/providers.
Issue being corrected: ______________________________________
Do not tune an idle reminder to fix all three.

3. EXPLICIT PROPOSED SETTINGS
UI: Assistants > saved assistant > Workflow > Call behavior > Messages and timeouts.
Idle message: _______________________________________________
Idle timeout (sec): ________ (zero disables idle prompts)
Max idle messages (total per session): _______________________
Maximum call length, reviewed separately: ____________________
Baseline values retained at: ________________________________

Fictional example, not a universal recommendation:
Idle message: Are you still with me? You can tell me the reference when you're ready.
Idle timeout: 25 seconds
Max idle messages: 2
Maximum call length: inspect actual admitted allowance separately.

4. LOGICAL TIMELINE, NOT A WALLCLOCK BENCHMARK
A. Latest relevant activity -> eligible quiet interval reaches 25 sec -> reminder1 requested.
B. Speech/activity settles -> another eligible quiet interval reaches 25 sec -> reminder2 requested.
C. Speech/activity settles -> later eligible quiet interval reaches 25 sec -> request idle-timeout finish, no reminder 3.

Activity restarts the timer. It does NOT replenish the total reminder allowance.
Caller returns after reminder1: normal conversation may resume; one reminder remains.
Caller returns after reminder2: normal conversation may resume; no unused reminders remain.
Do not add 25 + 25 + 25 and promise an exact 75-second call. Speech duration, activity, periodic checks, scheduling and earlier admitted limits affect actual time.
A finish request does not prove the caller heard an End call message.

5. PROPOSED ACCEPTANCE CASES
For every case fill: intended configuration | expected | observed | evidence | correction/owner | status.
Leave status UNTESTED until an authorized observation exists.

A. Caller says they are finding a reference, then stays quiet.
Expected: reminder after an eligible inactive interval, not during unrelated speech.
Observed: __________________________________________________
Evidence/time/route: _________________________________________
Correction/owner: __________________________ Status: UNTESTED

B. Caller continues speaking longer than the delay.
Expected: no idle reminder begins while caller is speaking.
Observed: __________________________________________________
Evidence/time/route: _________________________________________
Correction/owner: __________________________ Status: UNTESTED

C. Assistant gives a longer answer.
Expected: its own speaking state prevents an idle reminder over itself; finishing speech restarts activity timing.
Observed: __________________________________________________
Evidence/time/route: _________________________________________
Correction/owner: __________________________ Status: UNTESTED

D. Caller returns after first reminder and later goes quiet again.
Expected: activity restarts delay; only one total reminder remains in the two-reminder example.
Observed: __________________________________________________
Evidence/time/route: _________________________________________
Correction/owner: __________________________ Status: UNTESTED

E. Two reminders already used, then another eligible inactivity interval.
Expected: request idle-timeout finish rather than start reminder 3. Separately inspect actual disconnect/end reason and any audible farewell.
Observed: __________________________________________________
Evidence/time/route: _________________________________________
Correction/owner: __________________________ Status: UNTESTED

F. Explicit idle timeout zero.
Expected: no idle-reminder task starts for this configuration.
Observed: __________________________________________________
Evidence/time/route: _________________________________________
Correction/owner: __________________________ Status: UNTESTED

G. GPT Live with a saved positive idle timeout.
Expected: treated as incompatible; do not promise reminders. Review the explicit zero setting and broader mode requirements separately.
Observed: __________________________________________________
Evidence/time/route: _________________________________________
Correction/owner: __________________________ Status: UNTESTED

6. REVIEW AND RELEASE RECORD
Save changes is intentional; it can affect connected calls under legacy live settings.
Versioned telephone publication is separate from saving a draft.
Reviewed settings/version: ___________________________________
Authorized later test receipt: _______________________________
Actual publication action/version, if applicable: ____________
Actual route acceptance and limits: __________________________
Unresolved observations: ____________________________________
Owner/date for next review: __________________________________

Framework context: https://docs.livekit.io/agents/logic/turns/
Cost components: https://burki.dev/pricing
No field value, local logic test or worksheet alone proves caller-heard audio or telephone acceptance.