Background jobs
Seventeen workflows run behind the console: processing documents, generating and auditing Q&A, refining weak answers, indexing, rendering answer and service clips, seeding intent responses and delivering webhooks.
One engine behind every conversation
Servly separates what your persona knows and does from how it appears. The surfaces change; the knowledge, services and rules stay in one place.
Surfaces
Where people meet the persona
- Humanoid avatar
- Browser copilot
- Messaging (early access)
- Phone (early access)
- Mendix apps
Conversation agent
Decides what to do with every message
- Intent detection
- Clarifying questions
- Session memory
- Arabic and English, mid-conversation
- Hands over to a person
Knowledge and workflows
What the persona knows and can do
- Knowledge bases with their own index
- Audited Q&A
- Knowledge gaps
- Visual e-service flows
- Validation and OTP
Providers
Swappable, per knowledge base
- Language model
- Speech to text
- Text to speech
- Avatar and lipsync
Your systems
Where results land
- API calls from workflows
- Webhooks on completion
- REST API
- CSV and JSON exports
Workflow engine
Design e-services with steps, conditions, API calls and endings. The persona offers them in conversation, validates every answer against 34 rules, and sends completed submissions to your systems.
Knowledge and gaps
Documents become audited answers in a search index per knowledge base. Questions the persona can't answer become gaps your team closes once, for every surface.
An agent that uses tools, not just words
In agent mode the persona works through a small, audited set of tools. Every tool call stays inside your tenant, and the agent can only play media that belongs to you.
- Search answers
- Search services
- Offer a service
- Start, advance or cancel a service
- Play an answer or a service step
- Ask for clarification
- Remember and recall details
- Log a knowledge gap
Bring your own models and renderers
Connect the providers you already trust, test the connection from the console, and change them later without touching your knowledge or workflows.
- Language model
- OpenAI, Anthropic, Google, Custom endpoint
- Speech to text
- OpenAI, Google, Azure, Deepgram, ElevenLabs, Custom
- Text to speech
- ElevenLabs, OpenAI, Google, Azure, Custom
- Avatar and lipsync
- Built-in pipeline, Your renderer, No face: voice or text only
Run it with confidence
Evaluation
A test harness replays conversations and measures response latency, whether the right answer was played and how the agent handles services, detours and out-of-scope questions, so changes are checked before they ship.
Analytics
Widget usage by domain, answer performance and coverage, workflow funnels, and the knowledge gaps your customers hit most.
Multi-tenant by design
Each organization's data, files and search indexes are separated. Resellers get their own accounts, a sales toolkit and a white-label console.
See the console
We'll walk you through knowledge, workflows, gaps and analytics on a live tenant.