Supported LLMs
Runs on a model-agnostic architecture. Supports multiple LLM providers, including OpenAI and Azure OpenAI, as well as CBOT's own model, CBOT LLM.
Documentation
CBOT Platform Capabilities
CBOT is a Voice AI and Enterprise AI platform developed in Turkey that can run in the institution's own data center with on-premise deployment. This page lists the platform's capabilities by category, each linked to the page where it is explained in detail.
AI Engine
Bring Your Own Model
The institution connects its own model, adopts new models and switches providers without redeploying.
Models Running In-House
LLM inference, speech recognition, speech synthesis, RAG, vector store and the AI Agent runtime all run in the institution's own data center.
Knowledge Base and RAG
AI Agents generate answers from the institution's own documents through RAG and a vector store.
Function Calling and API Actions
During the conversation, the AI Agent connects to the institution's systems and completes the transaction on the spot.
Guardrails and Topic Boundaries
Business rules work as guardrails. Model access, routing and topic boundaries are defined and tested before going live.
Multi-Agent Orchestration
Specialized AI Agents hand off to one another under a single orchestrator, each taking on the work it does best.
Human-in-the-Loop Control
Sensitive decisions go to a person for approval. Once approved, the flow continues where it left off.
Confidence-Based Routing
Every step carries a confidence score. The AI Agent executes when confident, falls back to rules when unsure or hands the work to a person.
Smart Model Routing and Fallback
The best model is chosen for each task based on accuracy, latency, cost and data sensitivity. If a model slows down or becomes unavailable, it switches to an alternative automatically.
Separate Knowledge Domains
Knowledge is separated by product, region or business unit. Each AI Agent works only with the information it should see.
Purpose-Built CBOT LLM Models
Separate models are available for understanding, generation, summarization, classification and extraction. Models are fine-tuned to the institution's terminology.
No Dependency on External Model APIs
CBOT models run without depending on an external commercial model provider.
Enterprise Language Understanding (NLU)
Intents and entities are understood in a controlled way in structured enterprise scenarios.
CBOT Core
Intent Recognition and Entity Extraction
CBOT's own NLP and NLU recognize the user's intent and extract entities from the message.
Structured Dialog Management
Dialogs are managed with structured flows and rule-based business logic.
Classifier Training
The intent classifier is trained on the institution's own data.
Separate Test and Live Environments
Changes are tried in a test environment first, then moved to live.
Web Widget and Custom Theme
The chat widget added to the website is customized to the institution's brand.
Conversation Logs and Statistics Dashboard
Conversation logs, the most frequent intents and usage statistics are tracked in a single dashboard.
Surveys, Synonym Dictionary and Typo Tolerance
Surveys run at the end of conversations. A synonym dictionary and typo tolerance improve understanding accuracy.
CBOT AIFlow
No-Code Flow Design
Business teams design and update flows without code in CBOT AIFlow's visual environment.
Goal-Based AI Agent Design
Instead of a rigid script, the desired outcome is defined. The AI Agent interprets the goal, plans the steps and adapts as the request evolves.
Conditional and Parallel Execution
The AI Agent branches based on real conditions and runs independent steps at the same time.
Reusable Flow Components
A step is built once and reused across all flows.
Hybrid Execution
Fixed rules run on deterministic logic, complex work runs on reasoning models. The institution decides where each step runs.
Execution Logs and Operational Monitoring
Every step of every flow is logged, and operations are monitored in real time.
Run History and Retry
Every flow run is kept in history, and a failed run can be retried.
Credential and Permission Management
Integration credentials and variables are stored securely, and user permissions are granted per project.
Speech Technology
Speech Recognition (STT)
CBOT's own speech recognition runs in real time, is deeply optimized for Turkish and supports multilingual models.
Speech Synthesis (TTS)
CBOT's own speech synthesis engine produces natural voice with low latency. Supports SSML, pronunciation dictionaries, intonation and speed control.
Custom Lexicon and Pronunciation
Custom pronunciations are defined for brand and product names, and pronunciation is managed through a dictionary.
Speech-to-Speech
Spoken input, AI reasoning and spoken response are processed in a single flow. Fewer intermediate conversions mean lower latency.
Custom Brand Voice
A custom voice is developed to fit the institution's identity. Voice cloning is supported.
Low Latency
End-to-end response time stays under 1 second.
Language Support
Turkish Language Support
Developed Turkish-first. All products serve in Turkish.
48 Languages
Voice and text conversations are handled in 48 languages.
Conversation Capabilities
Handover to a Live Agent
The AI Agent hands the conversation to an agent with its full context. The handover runs through CBOT Live Chat.
Outbound Calls and Campaigns
Payment reminders, collections, notifications, appointment reminders, surveys and campaign calls are handled by voice AI agents.
Sentiment and Intent Analysis
KALYA tracks the trend in customer satisfaction and raises an alert when it drops. Intent performance is reported.
Answering Machine Detection
Answering machine detection for outbound calls is developed within the project scope, based on the institution's needs.
Interaction Memory and Cross-Channel Continuity
The AI Agent remembers what was said throughout the process. A process that starts on voice continues in chat or on the web without losing any detail.
Silent Transfer and Collaboration Mode
The conversation context is passed to the agent in the background. After the handover, the AI keeps supporting the agent with context and suggestions.
AI Assistant for Agents
Summarizes the ongoing conversation, drafts replies and suggests the next step.
Live Chat Management
Conversations are queued by request type and assigned to an agent with the right skills. The institution sets how many chats each agent can handle at once.
Telephony
SIP Trunk Support
Integrates with SIP infrastructure, IVR environments and contact center platforms. IVR, SIP and MRCP connections are supported.
PBX and CCaaS Integrations
Works over SIP with contact center platforms such as Genesys, Amazon Connect, NICE CXone, Avaya, Cisco, Twilio, Asterisk and Five9.
Concurrent Call Capacity
Concurrent call capacity scales with the deployment configuration and the carrier infrastructure.
Channels
WhatsApp and Digital Channels
The same AI Agent works across web, WhatsApp, Instagram, email and enterprise channels.
SMS
SMS is supported as a channel in the integration ecosystem.
Microsoft Teams and Facebook Messenger
In addition to WhatsApp and Instagram, it also works on Microsoft Teams and Facebook Messenger.
Integration and Development
Webhooks and Event Streams
Events are delivered to the institution's systems through webhooks and event streams.
Ready-Made Integrations
Ready-made integrations are available for CRM, case management, payments and other enterprise systems.
MCP Tools and Custom Connectors
Approved tools and enterprise services are exposed to AI Agents through MCP-based connections. Custom connectors are built for databases and legacy systems.
Versioning and Rollback
Every prompt, tool and flow is versioned. When needed, you can roll back instantly to the last stable version.
Security and Compliance
KVKK Compliance
Complies with KVKK, Turkey's personal data protection law. With the on-premise deployment option, data stays fully under the institution's control.
Personal Data Anonymization
Anonymization, retention and deletion policies are defined as part of data governance.
Data Retention Control
The institution sets its own retention, anonymization, deletion and access policies.
Single Sign-On (SSO)
SSO and multi-factor authentication are supported.
Role-Based Access
Works with role-based authorization. Data is encrypted in transit and at rest.
Audit Logs
Call recordings, transcripts, model decisions and workflow steps are logged and auditable.
Voice Biometrics
The caller's identity is verified from their voice.
GDPR Compliance
Personal data is processed in line with GDPR.
ISO 27001
Information security management follows the ISO 27001 standard.
PCI DSS
Payment card data is processed in line with PCI DSS requirements.
Deployment and Data Residency
Local, In-Country Solution
CBOT's speech recognition, speech synthesis and LLM technologies are developed in Turkey. The entire platform runs on the institution's own infrastructure with on-premise deployment.
Deployment Models
Cloud, hybrid and on-premise deployment options are available.
Data Residency
With on-premise deployment, data never leaves the institution's own data center.
SaaS and Private Cloud
On the cloud side, SaaS and dedicated private cloud options are also available.
Container-Based Architecture
The platform is packaged for Docker, Kubernetes and OpenShift. Inference, speech and orchestration services scale separately with the load.
Scaling and High Availability
Services scale out horizontally with demand, and warm model containers keep the first request fast. Clustering, failover and disaster recovery architectures are supported.
Operations and Monitoring
Analytics Dashboard
Call volume, completion, containment, fallback, escalation, average duration and satisfaction are tracked in a single dashboard.
Call Recording and Transcripts
Call recordings and transcripts are stored and reviewed in the approved deployment environment.
Automated Quality Assessment
KALYA evaluates conversations automatically and turns deviations and satisfaction drops into alerts.
Containment Reporting
Containment, fallback and escalation rates are reported in the dashboard.
Failure Analysis and Interaction Review
Low-confidence interactions, missed intents and escalation reasons are analyzed. Each conversation's transcript, confidence data and execution details can be opened.
Digital Employees
Ready-Made Digital Employees
Role-based Digital Employees: TAHSİLDAR for collections, GÜVEN for claims, ADİL for emergencies and KALYA for quality management.
Coaching Suggestions and Micro-Training
KALYA turns the gaps it detects into actionable suggestions for team leads and recommends short, targeted training when needed.
Document and Visual Damage Checks
GÜVEN reads vehicle registrations, driver's licenses, accident reports and damage photos, checks their consistency and routes risky cases to manual review.
Automatic Activation During Earthquakes
ADİL monitors official sources. When an earthquake exceeds a set magnitude, it sends employees a safety check over WhatsApp.
Services and Company
Turkish and English Support
The Istanbul-based team provides support in Turkish and English.
Deployment and Integration Services
Deployment and integration are delivered by CBOT's enterprise delivery team.
Office in Turkey
Istanbul Technopark ARGEM Building, Istanbul University Avcılar Campus.
References in Turkey
More than 50 published customer stories: Garanti BBVA, İşbank, QNB Finansbank, VakıfBank, Ziraat, Halkbank, Türk Telekom, Getir, Teknosa, Pegasus and more.
Time to Go Live
A typical project goes live within 2 to 4 weeks.
Service Level Agreement (SLA)
Uptime and support terms are guaranteed by an SLA in the enterprise contract.
This list is updated regularly. For deployment terms, certifications and contract details, talk to our team.
Talk to the CBOT team