Technical specification
System requirements
| Minimum (test) | Recommended (workstation) | Enterprise (server) | |
|---|---|---|---|
| Processor | 2 cores (x86-64 or ARM64) | 8 cores (x86-64) | 16+ cores (x86-64) |
| Memory | 4 GB | 16 GB | 32+ GB |
| Graphics card¹ | 6 GB VRAM | 12 GB VRAM | 64 GB VRAM |
| Storage | 10 GB free | 100 GB (SSD recommended) | 500+ GB (SSD/NVMe required) |
| Network | internet for the installation | stable, 100 Mbit/s and up | dedicated network card, 1 Gbit/s and up |
¹ The graphics card is optional. It determines how capable the local models can be – without a GPU, local models stay small. Cloud models remain accessible at any time, so demanding tasks are handled even without a graphics card of your own.
Operating systems
| Operating system | Versions | Support² |
|---|---|---|
| Windows | 10, 11 and newer | fully supported |
| Ubuntu | 20.04 LTS, 22.04 LTS, 24.04 LTS | experimental |
| RHEL / CentOS | 8, 9 | experimental |
| Debian | 11 Bullseye, 12 Bookworm | experimental |
| Fedora | 38 and newer | experimental |
| macOS | 12 Monterey and newer | experimental |
² "Experimental" means: binaries are available, but Windows is the platform for which we guarantee full operation and which we publish directly. For access to the Linux or macOS binaries, simply contact us.
Prerequisites on the machine
| Software | From version | What for |
|---|---|---|
| GPU drivers (NVIDIA) | – | Monitoring and use of the graphics card. Optional. |
| Operating-system keychain | – | Encrypted storage of credentials. Present on Windows and macOS; on Linux the Secret Service needs to be installed. |
Operation and interfaces
| Access | What for |
|---|---|
| Dashboard in the browser (local, port 8080) | Monitoring, costs, agent management, configuration |
| Chat assistant (Open WebUI) | The familiar chat interface for employees: every approved model – local or cloud – is available in the same window, while policies and logging run through DAOS |
| Command line, more than 60 commands | Full administration without a UI, scriptable and automatable |
By default, the dashboard and the interfaces are reachable only on the local machine. Exposing them on the network takes an explicit configuration step.
Models and providers
| Provider | Operation | Access |
|---|---|---|
| Ollama | local, on your own network | no access key needed |
| Anthropic Claude | cloud | API key |
| Google Vertex AI (Gemini) | cloud | service account as a JSON file; regions selectable |
| OpenAI | cloud | API key |
| Other providers with an OpenAI-compatible interface | cloud or local | address and, if required, a key |
Security and encryption
| Purpose | Method |
|---|---|
| Secrets and credentials | Kyber-1024 (NIST PQC level 5) combined with AES-256-GCM |
| Connections between the components | TLS 1.3 |
| Local database | encrypted on disk |
| Sign-in | token, valid for 24 hours; access without signing in is not possible |
| Permissions | role-based, seven predefined roles |
Security tiers
Five tiers, ascending by protection need. Only the two lowest may leave the building, that is, be passed on to external models:
| Tier | Meaning |
|---|---|
| Public | public, no protection need |
| Confidential | internal, normal protection need |
| Strictly Confidential | strictly confidential |
| Secret | secret |
| Private | highest tier |
Retention, limits, and budgets
- Metrics: 24 hours of history (default).
- Logs: six months directly accessible, and beyond that up to five years in the archive. An archive retrieval takes up to four hours.
- Budget periods: hourly, daily, weekly, monthly, or across the entire term.
- Budget warnings: at 50, 75, 90, and 100 percent of the limit.
- Responses when exceeded: warn only, pause the agent, stop the agent, or block new operations.
- Default limits: up to 100 agents per installation, up to 4 processor cores and 8 GB of memory per agent – all adjustable.
Compliance
DAOS is aligned with the following requirements; ISO 27001 certification is intended but has not been granted.
- EU AI Act, Article 12: requests and responses are logged encrypted and can be evidenced across the entire retention period.
- GDPR: data storage and processing logs are designed for it; for cloud components a data processing agreement (DPA) is concluded as standard.
- NIST Post-Quantum Cryptography: Kyber-1024 at level 5.
- Provider independence: technological independence from the large AI providers is contractually assured.
Licence model
Software licences by subscription
- DAOS Client License – billed per machine and month. Access for employees to steer AI agents and build workflows at their own workstation.
- DAOS Enterprise Hub – billed per server and month. The central control centre for governance, policies, user permissions, and running local models on the company network – as an on-premise solution.
Usage-based services
- Your own API keys (BYOK): free of charge. You pay only what the model provider charges you.
- Managed Inference via Dromni: official provider prices plus a 2 % handling fee.
- Log storage: directly accessible for up to six months, archived for up to five years – priced by storage volume.
Concrete amounts depend on the number of workstations, the term, and the scope. Book a quote call to receive your individual offer.
Status of this information
All figures on this page apply to version 0.1.1 (January 2026) and may change with future versions.
A binding breakdown for your environment – number of workstations, models, retention periods – comes out of the quote call; the path there is described under installation and start.