Business
RecommendedBest for: Teams currently using OpenAI, Gemini, Claude in their business processes.
Platform: 2 x RTX 5090 GPUs
Hardware cost: $20,000 once-off
Software cost: $10,000 pa
5-20 users
Private AI for your organisation
Deploy secure, on-premise AI for chat, research, coding, and AI agents without sending company data to cloud AI providers.
Why on-premise AI?
For organisations already using AI, the question is no longer whether AI is useful. The question is how to use it securely.
| Requirement | FirstStep On-Premise | Cloud Providers (OpenAI, Gemini, Claude) |
|---|---|---|
| Protect company IP | IP remains inside your infrastructure | Sensitive prompts leave your network |
| Keep customer data private | Customer data stays on-premise | Data is sent to external services |
| No runaway AI costs | Predictable platform costs without usage-based billing | Usage-based billing can spike unexpectedly |
| Unlimited AI usage | Unlimited usage within your private infrastructure | Usage limits and billing thresholds apply |
| Maintain governance and control | Central policy, access control, and audit logs | Limited governance across external providers |
Built for Compliance
Compliance
Stop sensitive data from leaving the organisation. Give teams a secure AI platform they are allowed to use.
IT
Deploy a managed AI platform without building and maintaining the full AI stack internally.
Finance
Replace uncontrolled AI subscriptions with one predictable platform and subscription.
Why organizations choose FirstStep
| Capability | FirstStep AI Systems | ChatGPT / Gemini | Private AWS Cloud | In-house Build |
|---|---|---|---|---|
| Cloud API | No | Yes | Yes | No |
| Models Available | Open Source | Vendor Controlled | Limited | Open Source |
| Latest LLM models | Auto-updated | Auto-updated | Manual | Manual |
| Performance | Enterprise | Enterprise | Enterprise | Hardware Dependent |
| Time to deploy | 1-2 weeks | 1-2 hours | 8-16 weeks | 12-18 months |
| Team Governance | Fully Auditable | No | Fully Auditable | Limited |
| IP exposure to 3rd parties | No Risk | High Risk | High Risk | No Risk |
| Customer Data shared | No Risk | High Risk | High Risk | No Risk |
| Air-gapped deployment | Yes | No | No | Yes |
| Token limits | No token caps | Yes | Yes | No token caps |
| Cost overrun risk | No Risk | High Risk | High Risk | No Risk |
| Integration to your servers | Yes | No | No | Yes |
| Integration to GIT repos | Yes | No | No | Yes |
One Managed Software Platform
Desktop Application
One desktop application (Windows, MacOS or Linux)
LLM Inference Engine Layer
High-performance serving of your models.
Linux Kernels and GPU Drivers
Our hardware is optimized for fast LLM inference with large LLM models.
API Gateway
Access your secure AI Workstation remotely.
Software Subscription
You receive the latest LLM models as updates.
Easy Onboarding
Configure Codex / VS Code to use on-prem LLMs.
Deployment options
Available as software-only, or as a complete AI Workstation solution.
Enterprise
Best for: Large organisations with multiple departments and offices.
Platform: Multiple AI Workstations
Hardware cost: $20,000 once-off per workstation
Software cost: $10,000 pa per workstation
20+ users
Software Only
Best for: Organisations with existing AI workstation hardware.
Platform: Your hardware
Hardware cost: Your own hardware
Software cost: $10,000 pa
Hardware dependent
FirstStep delivers a practical, affordable AI Workstations for businesses. Blackwell server racks available on request.
Financial Case Study
Hardware
6 Month ROI
On-premise hardware costs the equivalent of 6 months of cloud GPU rental (AWS, Azure).
Software
90% Lower Cost
$10,000/year instead of a typical $80,000 to $120,000 AI infrastructure engineer.
Security
100% Private
Your models and data remain on-premise, under your control.
Software Comparison
| Vendor | AI Models | Hardware Cost | Annual Cost (10 people) | Security |
|---|---|---|---|---|
| OpenAI | Private | $0 | $6,000–$60,000 variable | None (data shared publicly) |
| Claude | Private | $0 | $6,000–$60,000 variable | None (data shared publicly) |
| Gemini | Private | $0 | $6,000–$60,000 variable | None (data shared publicly) |
| FirstStep One | Open Source | $20,000 | $10,000 | Private, on-premise |
Hardware Comparison
| Hardware | Max Model Size | Speed | Hardware Cost (USD) | Best For |
|---|---|---|---|---|
| FirstStep AI Workstation | đĄ Medium (35B) | đĸ Fast | đĸ Low ($20,000) | Business (5-20 users) â |
| NVIDIA Gaming PC | đ´ Small (10B) | đĄ Medium | đĸ Low (~$3,000) | Personal |
| Apple MacBook Pro M4 Max | đĸ Large (55B) | đ´ Slow | đĸ Low (~$5,000) | Personal (developer) |
| H100 AI Server | đĄ Medium (35B) | đĩ Very fast | đ High (~$100,000) | Enterprise |
| B200 AI Server | đĩ Very large (80B) | đĩ Very fast | đ High (~$150,000) | Enterprise |
| Phone (Galaxy S25 Ultra) | đ´ Small (4B) | đ´ Slow | đĩ Very low (~$1,200) | Personal |
| DGX B200 (8Ã B200) | đĩ Very large (670B) | đĩ Very fast | đ´ Very high (~$600,000) | Hyperscale |
AI Platform Subscription
Stay current with the rapidly evolving AI landscape without managing models, updates, and compatibility yourself.
Model updates and access
One-click access to newly supported AI models, with curated recommendations built for enterprise use.
Automatic platform maintenance
Automatic platform updates plus security fixes and compatibility testing to keep your deployment stable.
AI Intelligence News Alerts
Bi-weekly AI intelligence briefings help teams stay current without constant market monitoring.
Priority support
Priority support ensures fast response when teams need deployment, model, or workflow guidance.
Useful for every department
Give every department a secure AI tool they are allowed to use.
Sales
Research prospects, analyse competitors, prepare proposals, write outreach messages, and respond to tenders.
Development
Use coding assistants for debugging, code review, documentation, architecture support, and test generation.
HR
Draft job descriptions, prepare interview questions, create onboarding material, and summarise policies.
Finance
Analyse spreadsheets, summarise reports, explain numbers, draft budget narratives, and review financial documents.
Projects
Summarise meetings, generate action lists, draft project plans, manage risks, and create status reports.
Marketing
Generate campaigns, research competitors, write content, create SEO briefs, and draft social posts.
Legal and Compliance
Review policies, compare documents, summarise contracts, and prepare compliance drafts.
Executive
Prepare market research, strategy notes, board papers, decision support, and business analysis.
Supported Models
FirstStep One now offers support for the following reasoning and coding LLM models:
Fast Reasoning Model
Fast, responsive AI for conversational agents and simple tasks.
Model
Gemma 2 9B
Vendor
Google
Google
Modality
Parameters9B
Size (VRAM)8.0 GB
Licenseđ Apache 2.0
FirstStep Benchmark Scores
50 input / 50 output tokens1,900 RPM < 1s (P95)
2K input / 200 output tokens100 RPM < 8s (P95)
Advanced Reasoning Model
Advanced reasoning and analysis for complex business questions.
Model
Gemma 4 12B-it
Vendor Google
Google
Modality
Parameters12B
Size (VRAM)—
Licenseđ Apache 2.0
FirstStep Benchmark Scores
50 input / 50 output tokens350 RPM < 4s (P95)
2K input / 200 output tokens350 RPM < 4s (P95)
Coding Model
Fast coding assistance for development tasks and software design.
Model
Qwen3-Coder 30B
Vendor
Qwen
Qwen
Modality
Parameters30.5B
Size (VRAM)18.7 GB
Licenseđ Apache 2.0
FirstStep Benchmark Scores
50 input / 50 output tokens60 RPM < 0.5s (P95)
2K input / 200 output tokens30 RPM < 8.0s (P95)
Image Generation Model
Fast image generation and editing for creative and visual workflows.
Model
FLUX.2-klein-4B
VendorBlack Forest Labs
Modality
Parameters4B
Size (VRAM)~13 GB
Licenseđ Apache 2.0
FirstStep Benchmark Scores
100 × 100~1.24s
512 × 512~2.93s
1024 × 1024~9.64s
Image Analysis Model
Fast multimodal analysis for images, documents and visual content.
Model
Qwen3-VL-8B
Vendor Qwen / Alibaba
Qwen / AlibabaModalityImage + Text → Text
Parameters8B
Size (VRAM)~6 GB+
Licenseđ Apache 2.0
FirstStep Benchmark Scores
100 × 100Coming soon…
512 × 512Coming soon…
1024 × 1024Coming soon…
Audio Transcription Model
Fast, multilingual speech-to-text for audio recordings.
Model
Whisper Large V3
VendorOpenAI
Modality
Parameters1.55B
Size (VRAM)~4 GB
License🔓 Apache 2.0
FirstStep Benchmark Scores
10s audio0.3s transcription
30s audio0.8s transcription
300s audio5.7s transcription
*RPM = requests per minute
New models are regularly tested and added, including models for image and audio processing.
Applications
FirstStep One brings essential AI and business applications together in one secure desktop application.
Technology Partners
Implementation Partners
