New: AI Agents 1.0
Create an AI Phone Assistant in 10 Minutes. With AI Agents, Carrier Service Factory 4 turns any phone number into an intelligent assistant that answers with a natural voice, knows your content and takes real action – see the overview below.
Security reviewed by AI
In addition to conventional testing and code review, Carrier Service Factory 4 is continuously checked for security using AI-based analysis, helping to identify and address potential issues early.
The table below gives a high-level overview of the features newly introduced with Carrier Service Factory 4 – the latest version.
| No. | Feature | Description |
|---|---|---|
| 1.0 | AI Agents – AI phone assistant | Create an AI phone assistant in 10 minutes: it answers calls with a natural voice, uses your website, documents and tables as knowledge, transfers calls to the right destination, books appointments in built-in calendars (also bookable online via a link or website snippet), calls your own functions, sends e-mails and SMS, verifies callers and hands over to existing services whenever needed. See the detailed overview below. |
| 3.1.1 | Latest frameworks for backend and frontend | Completely rebuilt on the latest .NET generation for a fast, modern and future-proof platform. |
| 3.1.2 | Real-Time Text (RTT, RFC 4103) | Real-Time Text is supported and can be switched on per service with a single option, so text is transmitted in real time alongside voice. |
| 3.2.1 | Advanced HTTP integration | A powerful HTTP integration to connect flows to any web service, with a built-in AI assistant that turns plain-language descriptions into ready-to-use logic and direct access to Dynamic Tables. |
| 3.2.2 | Dynamic Table delta updates | Dynamic Tables save only what actually changed, so very large tables stay fast and several people can edit them at the same time – both in the interface and via the Management API. |
| 3.2.3 | New Flow Editor | A completely redeveloped, web-based flow editor: drag & drop of dozens of blocks, undo/redo, grouping, zoom, reusable ports, parameters and variables, plus embedded help for every block. |
| 3.2.4 | OpenAPI import – whole APIs in one step | Paste the link to an API’s OpenAPI/Swagger description, tick the operations you need – and ready-to-use HTTP integrations are generated automatically, including URL templates, request bodies and encrypted credentials. The same import lets the AI phone assistant call entire APIs as actions. |
| 3.3.1 | Trigger Flow API | Launch outbound calls and campaigns from your own systems and receive their results back into the flow – a dedicated API with full OpenAPI/Swagger documentation and per-service API keys. |
| 3.3.2 | Resources Management API | Manage resource data such as Dynamic Tables and phone-number lists directly from your systems through a documented REST API secured with API keys. |
| 3.4.1 | Multi-factor authentication | Secure sign-in with multi-factor authentication based on modern identity standards; administrators can enforce MFA for their users. |
| 3.4.2 | Circuit routing & overflow protection | Automatic protection against routing loops and call floods: calls that exceed a configurable threshold are safely rejected. Flexible modes cover both classic loop protection and calls-per-second (CAPS) limiting, can be scoped to selected customer profiles, and apply instantly without a restart. |
| 3.5 | AI tools for service evaluation | A range of AI tools for building and evaluating services – including AI resources for leading providers and AI assistants embedded directly in the platform. |
| 4.1 | Media Path Optimization on by default | Efficient voice paths thanks to media path optimization enabled by default, adjustable per call when needed. |
| 4.1a | Codec renegotiation | Fine-grained control over media path optimization and codec renegotiation to avoid one-way-audio situations in demanding interconnect scenarios. |
| 4.2 | Background announcements | Play a looping announcement in the background while the flow keeps working – for example a "please wait" message during a lookup – and stop it at any time. |
| 4.3 | Configurable caller ID in outbound calls | Set the displayed caller ID (A-number) for an outbound call directly in the call block, with support for variables and parameters. |
| 4.4 | Dedicated permissions for power features | Powerful, script-capable blocks and scripting features are protected by their own permissions, so only authorized users can create or edit them. |
| 4.5 | Voice input with audio opt-in | Callers can respond by speaking or by pressing a key – whichever comes first – combining real-time speech recognition with classic key input in a single step. |
| 4.6 | Versioning of flows (incl. parameters) | Every save creates a version snapshot, including parameters. Compare any two versions in a clear diff view and restore an earlier state with one click; retention is configurable. |
| 5.1.1 | Dark theme & domain themes | Dark-mode toggle plus per-domain white-label theming with your own logo and colors. |
| 5.1.2 | Home / dashboard page | A landing page with call statistics, profile overview and quick access. |
| 5.1.3 | Solutions menu – fast loading | The solutions list loads on demand with server-side search and filtering, staying fast even with very many services. |
| 5.1.4 | Resources menu – fast loading | The resources list loads on demand with server-side search, type filter and sorting across all resource types. |
| 5.1.5 | AI Assistant chats | A full AI chat with saved history and auto-generated titles; private by default and optionally shareable read-only within a customer profile. |
| 5.1.6 | Documentation portal | In-app documentation with an embedded AI help chat. |
| 5.1.7 | FAQ collection | A curated FAQ covering routing/SIP, audio, APIs and more, in German and English. |
| 5.2.1 | Telemetry targets (Grafana / OTLP) | Open telemetry targets configurable per system and per call for live monitoring, for example with Grafana. |
| 5.2.2 | Usage statistics | Insights into how numbers, blocks and resources are used over time. |
| 5.3.1 | Customer profiles (multi-tenant) | Multi-tenant customer profiles with hierarchy and inherited visibility, with a profile selector on every page. |
| 5.4.1 | Tenant-scoped audit log | A per-tenant audit log so administrators only see their own tenant's activity. |
| 5.4.2 | Secrets encrypted at rest | Keys, passwords and other secrets are stored encrypted and are never shown again once saved. |
| 5.4.3 | SSRF protection for outbound HTTP | Outbound HTTP calls from flows are restricted to safe, public targets; internal destinations can be explicitly allow-listed. |
| 5.4.4 | Sandboxed scripting engine | Scripting runs in a safe sandbox with no access to the host system. |
| 5.4.5 | Stable authentication across servers | Authentication stays valid seamlessly across all servers and restarts. |
| 5.4.6 | Spoof-resistant client-IP handling | Client IP addresses are handled in a spoof-resistant way, trusting forwarding information only from the internal proxy. |
| 5.4.7 | Injection & data-leak hardening | Data handling is hardened against injection across queries, commands, file access and downloads, and error messages never expose internal details. |
| 5.4.8 | Per-IP request limit (rate limiting) | Built-in per-IP request limiting protects the APIs and web interface against request floods (DoS) – always on and configurable per surface. |
| 5.4.9 | Warning e-mails on protection triggers | Configurable warning e-mails notify administrators whenever a protection mechanism blocks something, throttled to avoid mail storms. |
| 5.4.10 | AI-assisted security review | The platform is reviewed for security with AI-based analysis in addition to conventional testing, helping to catch potential issues early. |
| 5.5.1 | New resource types | New resource types – including AI resources, enumerations and Dynamic Table definitions – with rich metadata. |
| 5.5.2 | "Used where?" usage tracking | See where each resource is used across all solutions. |
| 5.6.1 | Phrase hints for speech recognition | Phrase hints sourced from Dynamic Tables improve speech-recognition accuracy. |
AI Agents 1.0 – Create an AI Phone Assistant in 10 Minutes
Describe in plain language what your assistant should do, give it knowledge, pick a voice and assign a phone number – that’s it. No programming, no project: your AI phone assistant is live in about 10 minutes and answers calls around the clock.
Up and running in minutes
- Plain-language setup – instructions are written as simple text, and a built-in AI setup assistant helps you phrase them. Assign a phone number and the agent is on the line.
- Quick start with your website – enter a name and your website, and the homepage automatically becomes a knowledge source. Every new agent starts from a compact, proven default setup personalized with your company name; a phone number, a handover target and a voicemail address can be added right when creating it.
- Test before you launch – a built-in test chat lets you try the agent with its real knowledge and actions – handovers, bookings and e-mails are safely simulated – before it takes its first call.
Natural, multilingual conversations
- Natural voices – a choice of natural-sounding, multilingual voices from the latest real-time speech model, with smart end-of-speech detection and optional noise suppression; callers can interrupt the assistant at any time.
- Speaks your customers’ languages – natively multilingual with no language configuration at all: the assistant automatically answers in the caller’s language; an optional interpreter mode follows language switches mid-call and can even interpret between the caller and an employee.
Your knowledge, always current
- Every source you have – your website (with scheduled re-scraping), uploaded documents (PDF, HTML, Markdown, text), files from an FTP/SFTP directory, or whole data tables such as price lists and FAQs.
- Knows what it doesn’t know – questions the agent could not answer are collected as knowledge gaps, so you see exactly what content to add next.
Takes real action
- Transfers & handovers – to a single number, follow-me, parallel ringing or load-balancing targets, into any existing service, or to another AI agent; on handover, colleagues can receive the conversation so far by e-mail – shared across all handovers or with dedicated recipients per handover target.
- Business actions – calls your own functions (HTTP or scripts), sends e-mails and SMS, records voicemail messages, captures tickets and orders into tables and arranges automatic callbacks.
- Whole APIs at once (OpenAPI) – paste the link to an API description (OpenAPI/Swagger), tick the operations the assistant may use, adjust the wording if needed – done. The assistant calls the API on its own with the exact parameters from the description; write operations require explicit approval, credentials are stored encrypted, and every call runs under the platform’s security safeguards.
- Verified callers – the assistant verifies callers by an identifier and a secret (e.g. customer number and PIN) or automatically by their phone number against a stored list; sensitive actions and disclosures can be protected so they only run for verified callers – enforced server-side, and verified conversations are flagged in the call log.
- Appointments by phone & online – built-in calendars: the assistant offers free slots and books binding appointments, with appointment types of different durations, multiple places per slot, holiday block lists and minimum lead times. The same calendar can also be booked online: as a direct link or as an HTML snippet that embeds the booking page into your own website – visitors pick a date, the appointment duration and a free time slot and book bindingly. Bookings arrive as Outlook-compatible calendar invitations, calendars can be subscribed to in Outlook/Google, and automated reminder calls reduce no-shows. For security, an appointment can only be cancelled or moved by phone from the number that booked it; appointments booked online can be managed by phone the same way, via the phone number provided at booking.
Live view & quality
- Live monitor – watch running conversations with real-time transcripts and a live mood indicator; an AI suggestion summarizes the call and proposes the next step for an employee taking over.
- Measure and improve – an automatic summary of every conversation, caller ratings, automatic quality scoring by a second AI, topic tags with trend analysis, full-text search across all conversations and an A/B test for two instruction variants.
- Smart behavior – opening hours per action and transfer (outside those hours they are not even offered), a holiday/emergency mode (manual or driven by date lists), automatic escalation to a human when a caller is clearly frustrated, and recognition of returning callers.
Outbound, control & privacy
- Outbound calls too – the agent can also call out: reminder and campaign calls started via the API, with a per-call result report.
- Centrally managed & metered – usage is metered per customer profile, capacity limits can be set platform-wide and per customer profile, and automatic failover to a second region keeps the service available.
- Privacy by design – every conversation starts with a greeting that discloses the AI and, when transcripts are enabled, announces the recording – guaranteed by the platform; transcripts are switchable per agent with a configurable retention period and automatic deletion, and the caller’s phone number is never exposed to the AI.