Measure what matters, then make it better.
Shipping the agent is the start. Symphia gives you the console to see how it's really performing, a test suite to prove changes are safe, and an audit trail so every improvement is traceable.
The toolkit
See it, test it, trust the change
Everything you need to turn a live agent into a continuously improving one.
Insights console
KPIs on volume, resolution, and deflection in one place — see what customers ask and how well the agent answers.
Trends over time
Period-over-period movement on every metric, so you know whether a change made things better or worse.
Tool reliability
Track how often each function is called and how often it succeeds — catch a flaky integration before customers do.
Escalation reasons
Understand exactly why conversations reach a human, then close the gaps that drive avoidable handoffs.
Simulations
An agent-eval test suite runs scripted customers and grades the agent — so you ship changes without regressions.
Releases audit trail
A change log of who changed what and when, so every shift in behavior is traceable back to a decision.
Insights console
Know what's working — and what isn't
The Insights overview brings KPIs, period-over-period trends, top customer questions, and tool-usage reliability into one dashboard. Spot a dip in resolution, a spike in a single intent, or a tool that's failing, and act on it the same day.
- Resolution, deflection, and volume at a glance
- Period-over-period trends on every KPI
- Top questions and intents customers bring
- Tool-call reliability and escalation reasons
Simulations
Ship changes without regressions
Simulations is an agent-eval test suite. Scripted customers run against your agent and an LLM judge grades each transcript, with tag and tool-call checks confirming the agent did the right things. Run the suite before every change and catch regressions before customers do.
- Scripted users exercise real conversation paths
- An LLM judge grades each run against your criteria
- Tag and tool-call assertions verify correct behavior
- Tools mocked by default for fast, safe, repeatable runs
Releases
Every change, on the record
The Releases tab is an audit trail of how the agent has evolved — who changed which knowledge, tool, journey, or setting, and when. When behavior shifts, you can trace it straight back to the decision that caused it.
- Who changed what, and exactly when
- Covers knowledge, tools, journeys, and settings
- Tie a behavior change to the edit behind it
- Captured automatically — nothing to remember to log
Frequently asked questions
What does the Insights console show?
KPIs on volume, resolution, and deflection with period-over-period trends, plus top customer questions, tool-usage reliability, and the reasons conversations escalate — all in one dashboard.
What are simulations?
An agent-eval test suite. Scripted customers run against your agent and an LLM judge grades each transcript, with tag and tool-call checks confirming correct behavior, so you catch regressions before shipping.
Do simulations call my real systems?
Tools are mocked by default for fast, safe, repeatable runs. You opt specific simulations into live behavior only where it's appropriate.
How do I know what changed and why?
The Releases tab is an audit trail recording who changed which knowledge, tool, journey, or setting and when — so any shift in behavior is traceable to a decision.