AI to program the device .
Accelerate development of PLCs, custom robots, and industrial automation code by 5x.
Compatible with Siemens, B&R, CODESYS, Rockwell, Beckhoff, and 15+ ecosystems.
Completely hosted in Italy, the industrial IP remains with the company.
Siemens
B&R Automation
Codesis
Rockwell Automation
Beckhoff
Schneider Electric
Mitsubishi Electric
OMRON
Bosch Rexroth
ABB
Who
FANUC
Yaskawa
Universal Robots
Stäubli
Raspberry Pi
Arduino
Espressif ESP32
STM32 ST Micro
Adafruit
Realize industrial development and automation.
Not a generalist assistant. A specialized AI that understands ladder, structured text, GRAFCET, RAPID, KRL, ROS, OPC UA, and also MicroPython and CircuitPython for prototyping su Raspberry Pi e Arduino. Lavora in casa tua, sul tuo IP.
Write PLC code twice as fast.
Function Block generation in ladder and Structured Text, legacy code refactoring, guided debugging on machine logs.
- Function block generation for Siemens TIA, B&R Automation Studio, CODESYS
- Legacy STEP 7 / S5 → TIA conversion
- Test case automation with HMI and SCADA
- IEC 61131-3 Functional Document
Custom robotics, accelerated integration.
From study feasibility to trajectory code. Pick&place, palletizing, vision integration and force sensing.
- RAPID (ABB), KRL (KUKA), Karel (FANUC) road diameter generation
- Setting up a ROS 2 simulation Gazebo / Isaac Sim
- PID tuning, motion planning, Cartesian trajectory
- Cognex / Keyence / Halcon machine vision integration
Deliver more projects with the same team.
A productivity multiplier: your seniors focus on architecture and customer, not on writing repetitive code.
- Knowledge management of historical records
- Predictive maintenance with log machina
- Multilingual documents IT/EN/DE/FR/ES
- Automatic cost/time estimation based on customer specifications
Changing numbers of P&L.
Not a "AI revolutionizes everything" promise. Numbers measured on real tasks from an industrial automation team's workflow.
| Automation Engineer Task Tips | AI-free | CONSATE SOVEREIGN | Reset |
|---|---|---|---|
| Refactoring FB Legacy in TIA Portal (200 Ladders) | 2 Setman | 2 Days | −80% |
| Generating HMI test cases for new machines | 3 Days | 2 hours | −92% |
| Multilingual documentation IT/EN/DE | 1 Setimana | 1 hour | −98% |
| Automation Feasibility Study RFQ | 5 Days | 1 Day | −80% |
| Path Planning 6 Axis/Pick & Place Custom | 3 Days | 4 hours | −83% |
| Legacy STEP 7 / S5 → TIA Portal Refactoring | 4 Weeks | 5 Days | −75% |
| Arduino/Raspberry Pi IoT Prototype (Firmware + Dashboard) | 5 Days | 1 Day | −80% |
| Analysis logs by intermittent root cause | 2 Days | 1 hour | −94% |
From spec to integrated robotic cell.
We don't sell a code library — we sell a development companion that understands ABB, KUKA, FANUC, Yaskawa, ROS 2 and proprietary controller languaggi proprietari di ciascun controller.
Optimized trajectories in minutes
Cartesian and joint-space path generation on 6/7 axes with load and collision constraints.
End-to-end sensor integration
Cognex, Keyence, Halcon, ATI Force/Torque. Automatic calibration pipelines.
Change layout for each cycle
Stima Tact Time, Positioning Robot, Sikurezza ISO/TS 15066, Sequence Operator.
Simulation + virtual commissioning
Set up Gazebo / Isaac Sim / RobotStudio / RoboGuide.
Cell development time
From spec to final commissioning
marche supporter
ABB, KUKA, FANUC, Yaskawa Electric, UR, Stäubli
IP internal reuse
The AI learns from your libraries, not the other way around
Developers already speak the language.
SATE Sovereign natively integrates with OpenHands (the open-source autonomous agent) and templates Hermes from Nous Research. Your team keeps using the tools they know; only the endpoint changes, and everything stays in-house.
OpenHands (formerly OpenDevin) is the most adopted open-source AI agent for autonomous software engineering. It executes multi-step tasks in your sandbox: reads code, writes PRs, runs tests, navigates documentation. SATE Sovereign is the perfect backend: everything behind your VPN, no CLOUD Act.
- Drop-in using SATE Sovereign Base_url - no code changes
- In-house Docker sandbox, no prompts to US
- GitLab/Gitea/Bitbucket on-premises PR autonome
- Tickets → Store opening → PR → Review umana finale
The family Hermes (3 and 4) from Nous Research is one of the best for tool calling, function calling and agentic workflows. Open-weight, steerable, trained for complex instructions and structured output. On SATE Sovereign you can serve it privately with the same OpenAI-compatible API.
- Tools to invoke robust JSON schemas
- Deterministic output structure (XML, JSON, regular expressions)
- Maneuverable: No need to legitimately refuse the mission
- Open Weight: Bender's Nessun Lock-in
The cluster learns from your team. Every night.
SATE Sovereign include un Knowledge Distillation Layer: a process that every night captures successful interactions of the day, distills them into LoRA adapters for each specialist agent, and improves performance on your tasks — without ever modifying the Master Agent, which remains with reference of general knowledge.
Open weight Frontier optimized to industrial standards.
Selection of commercial licenses. Workflow includes La Famiglia Hermès for each agent.
| Model | Parameters | throughput | Optimal use case |
|---|---|---|---|
Sovereign LiteLama-3-70b · Ladder, ST, lib gen | 70B | ~45t/s | PLC code completion, log classification, technical document generation |
Sovereign Promixtral style 120B · Agent automation | 120B | ~30t/s | Refactoring complexity, multi-step debugging, and code review architecture |
Sovereign Hermes 4Nous-Hermes-4-70B · Agent Workflow | 70B | ~40t/s | Tool calls, function calls, output struts, OpenHands backend |
Sovereign Hermes XLnous-hermes-4-405b · Frontier Mystery · MASTER | 405B | ~9t/s | Master Agent: Teacher per Knowledge Distillation, General Administrator |
Sovereign RAGMistral-Large 123B · Dock Industrial | 123B | ~32t/s | RAG manual for Siemens/B&R/Rockwell, multilingual documentation |
Dedicated per ingest or per cluster.
Predictable costs in euros charged in Italy. Break-even points often occur on the first project.
- PLC & robotics code completion
- automatic documentation
- API OpenAI compatible
- Rate limit: 60 requests/min
- Legacy + agent refactoring
- OpenHands integration ready
- Distillation of Knowledge Notturna
- Hermes 4 70B included
- Hermes XL 405B Cammaster
- Systema architecture
- Context 128k token
- priority queue
- Capacity reservation 100%
- LoRA customized for your IP
- SLA 99.5% contract
- Custom DPA
Industrial IP never leaves the company.
American engineers may have bigger models. But if you send them your PLC code library built over 30 years, you're giving it away for free. SATE Sovereign is the European answer for those who can't afford to export their know-how.
Domand frequently.
What tech teams, CFOs, and DPOs want to know before signing.
Il modello vede i miei prompt? What services are available depending on the additional model of the client?
No, in two senses. First: prompts are not persisted on disk beyond the request duration — only anonymous metadata (tenant ID, timestamp, token count, latency) for billing and capacity planning. Second: your prompts are never used to train shared models. Nightly knowledge distillation only happens on interactions that your team explicitly marks as "success", and the resulting LoRA stays in your tenant — it's not shared with other customers or used as a public dataset.
What is the difference between Mistral, Aleph Alpha, and OpenAI?
Vs OpenAI/Anthropic: we are in Italy, under exclusive EU jurisdiction, no CLOUD Act. Prices 35-65% lower at equivalent tier. Vs Mistral / Aleph Alpha: they are great for 10k+ user enterprises; we are built for European SMEs that cannot afford suei contratti. We also offer nightly Knowledge Distillation Layer and fine-tuning on client IP — features that European sovereign hyperscalers don't have as standard. Vs all: our advantage is specialization in industrial code (PLC, robotics) and the option of full on-prem air-gapped.
Fully functional on-premises, but is there an air gap?
Yes. The Enterprise tier includes physical cluster provision in your data center with optional air-gapped configuration — no internet connection required after initial setup. You update the templates via offline download (signed files). PA, healthcare, defense and manufacturing with critical IP typically choose this configuration. Pricing starts at €2,400/month and includes hardware, maintenance, model updates and onsite support SATE.
How do you invoice? Italian VAT compliance?
Yes. SATE S. R. L. is registered in Italy (VAT 01132460476, REA PT-121825). Electronic invoicing via SDI, 22% VAT applied in Italy, reverse charge for EU customers with valid EU VAT ID (automatic VIES verification). Payment at 30/60 days by bank transfer, or SEPA Direct Debit for recurring customers. No Stripe US, no App Store subscription — all classic Italian B2B.
Do you support IEC 61131-3 v3? Vendor specific?
Yes for IEC 61131-3 v3 (LD, ST, FBD, IL, SFC). Models were exposed during fine-tuning to real code corpora for the main ecosystems: Siemens TIA Portal V18+ / SCL, B&R Automation Studio 4.x and 6.x, CODESYS V3.5 + Library, Rockwell Studio 5000, Beckhoff TwinCAT 3, Schneider Eco Trekchure. For robotics: rapid (ABB), KRL (WHO), Karel (FANUC), inform (Yaskawa), UR Script (Universal Robots), plus ROS 2 and MoveIt. If your stack has custom specifications, the nightly LoRA learns them from your projects.
Would you like to start Veromaintenance from the beginning?
For the Cloud tier (Light/Pro/Max): 24 hours to activate the API key and provide team credentials. Free 30-minute guided demo before signing to validate on your real code. For the Enterprise tier on-prem: typically 4-6 weeks from order to go-live, of which 2 of Strix Halo hardware lead time + 2-3 of setup, network, operational training. Private Beta Q3 2026 — select industry partners get free API credits for 3 months and 50% off the first 6 months.
Was Hermes 4 (Oltro model) a success?
Base model updates are transparent to the customer: internally tested on regression benchmarks (MMLU, HumanEval, MT-Bench) before deployment. If the model upgrade base breaks your LoRA, the system automatically rolls back to the previous model for your tenant until the LoRA is readjusted (1 night cycle). You have historical access to LoRAs from the last 30 versions — you can always go back to a version that worked.
Can I tweak my own code? Who owns it?
Yes, fine-tuning via LoRA is included from the Pro tier and above (nightly knowledge distillation), and is extended to manual fine-tuning on historical batches in the Enterprise tier. The resulting LoRA is owned by the customer: contractually the "model specialized on your IP" is yours, not ours. In case of cancellation, you will receive the LoRA weights (safetensors file) to take them elsewhere. The base model remains licensed by us, but exists free as open-weight (Llama, Hermes, Mistral) — you are never locked in.
Are SLAs guaranteed?
Tier Light/Pro/Max: 99% monthly (best effort, proportional credit refund for minutes of downtime above the threshold). Tier Enterprise: 99.5% contractual with penalties, phone support H24 5×8 (24×7 optional), guaranteed intervention time 4 hours during business hours. Scheduled maintenance 1× per month in an agreed nighttime window. Public status page on status.sate-sovereign.eu (coming soon) with real-time uptime and latency.
Does it work with our development tools? Plugins for TIA, Automation Studio, RobotStudio?
Three integration paths. (1) OpenAI-compatible API: use any tool that talks to OpenAI (Continue.dev, Cursor, Cline, Aider, GitHub Copilot Enterprise via proxy). (2) OpenHands installed in-house: autonomous agent operating on your internal Git repo. (3) Native plugins (in roadmap Q1-Q2 2027): extensions for TIA Portal, B&R Automation Studio, CODESYS, RobotStudio, RoboGuide. Meanwhile, VS Code integration (which many automation teams work with in parallel) is already complete via standard extensions.
Let's take a look at the code.
30-minute guided demo on a real case from your team. We show you the AI generating Function Blocks or robotics code from natural language, directly on your automation stack, without code leaving your machine.
Or write to info@sate.eu