Call recordings stays inside the organisation
Call recordings, transcripts, and evaluations never leave your organization. The entire platform, including the panel, CRM, speech recognition, and CALLAII language model, runs locally, fully offline if desired. A single workstation for small businesses, a redundant setup for enterprises.
The screen above is a representative example, not real deployment data.
From SMB to enterprise one architecture
The same Callmenta platform runs in three different configurations, scaled to your size.
SMB Workstation
The full Callmenta platform running on a single machine, with CALLAII 12B.
Small businesses
Single-Node Server
A single server for mid-to-large operations, with CALLAII 26B.
Mid-to-large operations
Enterprise redundant deployment
Redundant deployment is designed together during discovery.
Banking, insurance, public sector
16 GB on a single GPU is enough to start
A single NVIDIA card with 16 GB memory runs the entire platform, including CALLAII Edge Compact and speech recognition, for small businesses. As you grow, you move to a 48 GB card and redundant deployment.
Starter (SMB)
- 1× NVIDIA, 16 GB VRAM (for example RTX 4080 or RTX A4000)
- 32 GB DDR5 RAM
- 1 TB NVMe SSD
- Linux Ubuntu 22.04 / 24.04
CALLAII Edge Kompakt · capacity is measured according to your call volume before installation
Enterprise, redundant
- 48 GB GPU per node
- ECC RAM · NVMe RAID
- 10 GbE network
- Redundant deployment designed during discovery, offline option
CALLAII Edge Kurumsal · capacity is measured according to your call volume before installation
The model is selected to match the hardware during setup. On a card with 16 GB of memory, CALLAII 12B and speech recognition run together. CALLAII 26B requires a single 48 GB card or two 24 GB cards; 26B alone uses nearly the entire 24 GB card. Measurement (September 2026): on two 24 GB NVIDIA L4 cards (one running CALLAII 26B, the other running speech recognition), with real calls measured at average duration and two concurrent analyses, approximately 700 calls per day are analyzed under continuous load; speech recognition ran at approximately three times the speed of the audio.
Sized to your operation 3 packages
Hardware, licence and service level are set per package. Final pricing is built around your call volume and required precision.
Starter & Small Operation
Start with a single server. For small businesses and validation setups, with CALLAII Edge Kompakt.
- 1× NVIDIA GPU, 16 GB
- Callmenta platform + CALLAII 12B
- Speech recognition on the same card
- Standard support (business days)
- Updates via manual delivery or internal network
Mid-Scale
CALLAII Edge Enterprise for mid-sized and large operations.
- 1× 48 GB or 2× 24 GB NVIDIA GPU
- Callmenta platform + CALLAII 26B
- Redundant setup is designed together during discovery
- Priority support (8×5)
- License verified offline
Large Scale + Air-Gapped
Fully offline operation for critical infrastructure and regulated sectors.
- 48 GB GPU per node
- Redundant setup is designed together during discovery
- Fully air-gapped operation (offline license)
- 24/7 SLA + field engineer
- KVKK compliance report
Your records never crosses the perimeter
Call recordings, transcripts, and evaluations stay within your organization's boundaries. External connections only go to addresses you authorize.
Data sovereignty
Recordings, transcripts and evaluations stay inside the organisation. Compliant with KVKK, GDPR and sector regulators (BDDK, SPK, health authorities).
No cloud round-trip
Analysis runs on your own server; audio and text never leave the premises. Risky calls are escalated to the manager from within the organisation.
License portal
Track your license status, expiration date, and renewal requests in the license portal.
No change to your phone system
CALLAII Edge retrieves call recordings from your PBX recording folder. SIPREC support is set up as part of the project. All recording and analysis runs on your server. You do not change your operator or numbers.
CALLAII Pro is also an Edge deployment
Pro is the training layer added on top of the Edge layer. The model is trained on your own conversation data and runs on your GPU server. The training location is determined together during deployment.
The same server
If you already run Edge, Pro sits on top of it. No second stack is installed, the same nodes are used.
Trained on your data
Approved evaluations are converted into training data in ALPACA, ChatML, or JSONL formats. The training location and schedule are determined together during setup.
If you have no GPU server
We can also run your model on our hardware. In that case an activation fee replaces the installation fee and the monthly amount covers hosting.
Frequently asked questions about CALLAII Edge
What is CALLAII Edge?
CALLAII Edge is an on-premise deployment that runs the entire Callmenta platform (panel, call center, CRM, industry modules, and CALLAII AI) on your own server. Licenses determine which modules are active. Call recordings and transcripts never leave your organization.
Does my data leave the organization?
Call data does not leave. Recordings, transcripts, and evaluations remain on your infrastructure; no data is sent to Callmenta. If you enable integrations like WhatsApp or SMS, those messages go only to addresses you authorize. This suits the strictest data privacy scenarios under KVKK and GDPR.
Who is it for?
It suits anyone who needs privacy and wants to run Callmenta on their own infrastructure, including data-sensitive sectors such as banking, insurance, healthcare, and the public sector.
Does it run fully offline?
Yes, if required. Licenses are verified offline during installation, with no connection to Callmenta. Models are delivered separately during setup. Call data is never sent externally under any circumstances.
Does it offer the same capabilities as the cloud version?
Largely yes. All modules work on-premise, with licenses determining active modules. Legal research and some knowledge bases require a separate data package during installation. PBX records are retrieved from the recording folder. Internet-dependent integrations (WhatsApp, SMS, e-commerce) open only to external addresses you authorize.
Let's plan an Edge architecture review together
Let us plan the hardware list, the licence model and the rollout together.