Skip to main content
Agent maintenance guide Use this section when you have a live agent and need to make a targeted change — updating a topic, changing a voice, fixing a function, or adjusting a behavior. Each guide is scoped to a single task.
Drive most maintenance through Studio Assistant (Beta). Project admins on enterprise clusters can ask Studio Assistant in natural language to update topics, tweak flows, rewrite welcome messages, or fix a behavior that surfaced on a real call. Every change lands on a branch you review before merging — see the prompting guide for patterns that work well. Access is admin-only during Beta — see Role permissions.

​
Quick decision guide

I need to…Go hereTimeSkill level
Describe a change in natural languageStudio Assistant1-15 minNo code
Update FAQs or business infoFAQs10-30 minNo code
Sync external sources (websites, files)Sources5-15 minNo code
Change voice or fix pronunciationVoice and audio5-20 minNo code
Update or debug a functionFunction maintenance15-60 minRequires Python
Maintain multilingual agentsMulti-language updates30-90 minNo code
Check performance or fix issuesPerformance monitoring30-120 minNo code
Analyze conversations at scaleSmart Analyst – deep sampling up to 500 calls5-10 minNo code
Publish or rollback changesVersion management5-15 minNo code
Daily/weekly health checksHealth checks5-60 minNo code

​
Core maintenance tasks

Studio Assistant

Describe the change you want in plain language — Studio Assistant edits topics, flows, entities, and settings on a branch you review. Beta

FAQs

Edit existing answers or add new topics to guide your agent’s responses.

Sources

Sync and manage external sources like websites, files, and integrations.

Voice and audio

Update voice settings, fix pronunciations, and optimize audio quality.

Function maintenance

Update code, debug errors, and manage API integrations. Requires Python familiarity.

Multi-language updates

Maintain and optimize multilingual agent configurations.

Performance monitoring

Identify issues, track metrics, and improve agent performance.

Version management

Publish, promote, and rollback changes safely.

Health checks

Daily, weekly, and monthly maintenance checklists.

Smart Analyst

AI-powered conversation analysis – deep sampling analyzes up to 500 calls per query to surface patterns and insights.

​
Additional resources

Routing and handoffs

Update phone numbers, SIP routing, or call transfer logic.

QA and analytics

Listen to calls, assess quality, and review performance data.

Common issues

Troubleshoot frequent problems and find quick solutions.

Alerts API

Set up automated alerts for latency, errors, and call volume.

​
Typical workflows

​
Daily (5-10 minutes)

  1. Check overnight metrics
  2. Review recent calls
  3. Verify critical functions
  4. Address urgent issues
See: Health checks

​
Weekly (30-60 minutes)

  1. Run Smart Analyst deep sampling to surface the week’s trends and failure patterns across hundreds of conversations
  2. Update FAQs content based on Smart Analyst insights
  3. Sync Sources sources
  4. Fix identified issues
  5. Publish and promote changes
See: Smart Analyst, FAQs, Performance monitoring

​
Monthly (2-4 hours)

  1. Use Smart Analyst for a deep sampling review – analyze containment blockers, handoff reasons, and sentiment trends across the full month
  2. Optimize functions and performance
  3. Update voice and audio settings
  4. Address knowledge gaps identified by Smart Analyst
  5. Test and validate improvements
See: Smart Analyst, Voice and audio
When in doubt, use your environments carefully. Make any changes in the Sandbox, test thoroughly, and only then promote up to Live.
Last modified on June 18, 2026