# Platform Overview

## The Sovereign AI Laboratory

AVELIN is an **AI Laboratory** that builds frontier-class intelligence. We train and fuse models into specialist families — then deploy them inside your perimeter, on your GPUs, at half the cost of closed frontier providers.

We are not a chatbot wrapper, a token reseller, or a pass-through API gateway. AVELIN builds **Cross-Model MoE** — a trained Mixture-of-Experts meta-model architecture with difficulty-based cascading that synthesizes intelligence from the world's most capable open models. This is not routing. This is trained AI technology that produces outputs exceeding any single model on its own.

The result: intelligence that matches or exceeds frontier providers on independent benchmarks, at up to 85% lower cost, with the sovereignty and compliance controls that regulated enterprises require.

---

## Three Pillars

AVELIN operates across three integrated pillars:

### Pillar 1: AI Laboratory (Primary)

Like Mistral, we are fundamentally a research and engineering laboratory that builds AI models. Our core technology is **Cross-Model MoE** — a trained Mixture-of-Experts meta-model with difficulty-based cascading that fuses open models into **9 specialist model families**, each optimized for a distinct workload class:

- **Intelligence** — general reasoning, analysis, and complex problem-solving
- **Coding** — software engineering, architecture, and code generation
- **Agentic** — autonomous multi-step task execution and tool use

Each family contains multiple tiers (e.g., ultra, pro, fast) so enterprises can match intelligence level to task complexity — and never overpay for capability they don't need.

**We build and fuse models, we don't just route them.** Cross-Model MoE is a trained meta-model — with learned gating, specialist weighting, and difficulty-based cascading — that produces synthesized intelligence. New open-source models are evaluated and integrated within 48 hours, continuously improving the system without the $100M+ training budgets of closed providers.

### Pillar 2: Cloud Token Factory

Like Cohere, we operate an API-first serving infrastructure that delivers AVELIN's specialist models at scale — **up to 85% cheaper than frontier alternatives**.

- Drop-in compatible API (OpenAI-compatible endpoint)
- Automatic tier selection based on task complexity
- Sub-50ms routing overhead
- 99.9%+ uptime with automatic failover
- Real-time usage metering and spend controls

### Pillar 3: Enterprise Platform

Like Palantir, we deploy sovereign intelligence directly inside our customers' infrastructure. This is our **primary business model** — enterprises license AVELIN for deployment on their own hardware.

- **Sovereign GPU licensing** — deploy AVELIN on your own GPUs (annual licensing, priced per deployment — [request a POC](/poc#request))
- **On-premises deployment** — run entirely inside your own infrastructure
- **Air-gapped environments** — full operation without external network connectivity
- **Sovereign data residency** — data never leaves your approved perimeter
- **Compliance-ready** — designed for HIPAA, SOX, FedRAMP, and GDPR requirements
- **Role-based access governance** — policy controls enforced at every layer

---

## Cross-Model MoE: Trained Intelligence, Not Routing

Cross-Model MoE is the core technology that differentiates AVELIN from API aggregators and pass-through gateways.

| Aspect | API Aggregator / Gateway | AVELIN Cross-Model MoE |
|--------|--------------------------|------------------------|
| Architecture | Request forwarding | Trained MoE meta-model |
| Intelligence | None — passes through to providers | Synthesized — models fused with learned gating |
| Output quality | Limited to single model | Exceeds any individual model |
| Improvement | Dependent on upstream changes | Self-improving within 48 hours |
| Deployment | Cloud-only, data leaves perimeter | Sovereign — runs on your GPUs |
| Cost structure | Markup on provider pricing | Structurally cheaper architecture |

Cross-Model MoE uses **difficulty-based cascading**: each request is analyzed for complexity and routed through trained specialist layers. Simple queries use efficient pathways. Complex reasoning activates deeper expert fusion. The meta-model learns which specialists to weight for each task class — this is synthesis, not a gateway.

---

## Model Families & Performance

All AVELIN models are continuously evaluated against independent, third-party benchmarks — never self-graded. Evaluation results are published on the dedicated evaluation portal, eval.avelin.ai.

### Intelligence Family

General reasoning, complex analysis, and knowledge-intensive tasks.

| Model | Context | Max Output | Input $/M | Output $/M |
|-------|---------------|---------|------------|-----------|------------|
| intelligence-ultra | 55 | 256K | 65K | $2.50 | $12.50 |
| intelligence-pro | 53 | 256K | 65K | $1.40 | $4.40 |
| intelligence-fast | 47 | 256K | 65K | $0.30 | $1.20 |

**up to 85% cheaper than frontier alternatives** on equivalent intelligence workloads.

### Coding Family

Software engineering, code generation, architecture design, and debugging.

| Model | Context | Max Output | Input $/M | Output $/M |
|-------|---------------|---------|------------|-----------|------------|
| coding-ultra | 52 | 1M | 65K | $2.50 | $12.50 |
| coding-pro | 49 | 1M | 65K | $1.40 | $4.40 |
| coding-fast | 42 | 1M | 65K | $0.30 | $1.20 |

**up to 86% cheaper than frontier alternatives** on equivalent coding workloads.

### Agentic Family

Autonomous multi-step task execution, tool orchestration, and complex workflows.

| Model | Context | Max Output | Input $/M | Output $/M |
|-------|---------------|---------|------------|-----------|------------|
| agentic-ultra | 67 | 256K | 65K | $2.38 | $11.88 |
| agentic-pro | 67 | 256K | 65K | $1.26 | $3.96 |
| agentic-fast | 38 | 256K | 65K | $0.285 | $1.14 |

**up to 87% cheaper than frontier alternatives** on equivalent agentic workloads.

---

## Why the Cost Advantage Exists

Closed frontier providers spend $100M+ training a single foundation model, then charge premium prices to recoup that investment. AVELIN's Cross-Model MoE takes a fundamentally different approach:

- **Trained MoE meta-model** — we fuse and train specialist layers on top of open models, not just route between them
- **Difficulty-based cascading** — each request activates only the expert pathways it needs
- **Continuous integration** — new open-source advances are absorbed within 48 hours at negligible cost
- **No lock-in premium** — you're not paying for a provider's moat, you're paying for delivered intelligence

This is why AVELIN delivers frontier-competitive quality at up to 85% lower cost — structurally, not through subsidies.

---

## Business Model

### Enterprise Licensing (Primary)

Our primary business model is **sovereign GPU licensing** — enterprises deploy AVELIN on their own hardware, inside their own perimeter.

| | Details |
|---|---|
| **Pricing** | Annual per-GPU licensing, tailored per deployment — [request a POC](/poc#request) |
| **Hardware** | Customer provides GPUs |
| **Deployment** | On-premises, private cloud, or air-gapped |
| **Includes** | Full Cross-Model MoE portfolio, governance, continuous updates |

This model aligns incentives: we earn when enterprises succeed with sovereign AI, and they retain complete data control.

### Token Selling (Secondary)

For organizations that prefer usage-based consumption, we offer cloud API access:

| | Details |
|---|---|
| **Pricing** | $0.285–$12.50 per 1M tokens (varies by model tier) |
| **Billing** | Pay-per-token, no minimums |
| **Access** | Immediate via API, no infrastructure required |

---

## Product Portfolio

| Product | Role | Value |
|---------|------|-------|
| **AVELIN-GPT** | Unified conversational workspace | Faster daily execution with specialist assistants and workflow continuity |
| **AVELIN-API** | Cross-Model MoE serving layer | Drop-in API access to all specialist families with governance controls |
| **AVELIN-MCP** | Integration and action framework | Reliable insight-to-action execution across connected systems |
| **Y-RAY** | Intelligence and research engine | Evidence-backed analysis with multi-depth investigation modes |

### AVELIN-GPT

- Private conversational workspace for business users
- Classic, Memory, and Debate intelligence modes
- Specialist assistants powered by Cross-Model MoE families
- Multimodal interaction with files, media, and voice
- Cross-device continuity for uninterrupted workflows

### AVELIN-API

- OpenAI-compatible API for seamless integration
- Cross-Model MoE routing across all specialist families
- Automatic tier selection based on task complexity
- Token-level usage metering and spend governance
- Extensible guardrails and behavior controls

### AVELIN-MCP

- High-performance connector framework for action execution
- Controlled permission model for safe automation
- Cross-service workflows for communication, planning, and coordination
- Connector governance aligned to business risk posture

### Y-RAY

- Corporate Brain intelligence from internal knowledge assets
- Internal and external evidence fusion for stronger reasoning
- Multi-depth analysis modes:
  - **Alpha Scan** — fast signal detection
  - **Structural Breakdown** — strategic mapping
  - **Deep Diligence** — claim-level validation

---

## Why Organizations Choose AVELIN

### 1) An AI Laboratory, Not a Reseller

AVELIN builds trained AI models. Cross-Model MoE is proprietary technology — a trained meta-model that synthesizes intelligence from multiple specialists. You're partnering with a lab, not renting from an aggregator.

### 2) Frontier Intelligence at Half the Cost

Independently benchmarked. Third-party validated. Structurally cheaper — not subsidized.

| Workload | AVELIN Advantage |
|----------|-----------------|
| Intelligence | up to 85% cheaper than frontier alternatives |
| Coding | up to 86% cheaper than frontier alternatives |
| Agentic | up to 87% cheaper than frontier alternatives |

### 3) True Sovereignty — Your GPUs, Your Data

Deploy AVELIN on your own GPUs under an enterprise license. Your data never leaves your perimeter. Air-gapped operation for classified environments. This is sovereign intelligence.

### 4) Cross-Model MoE, Not Pass-Through

AVELIN doesn't proxy requests to third-party APIs. Cross-Model MoE synthesizes intelligence from multiple specialist models, delivering quality that exceeds any single model on its own. This is trained AI technology — not a gateway.

### 5) Continuously Improving

New open-source model advances are evaluated and integrated within 48 hours. Your intelligence improves continuously without migration projects or API changes.

### 6) Compliance-Ready Architecture

Designed for regulated environments: HIPAA, SOX, FedRAMP, and GDPR. Role-based access, audit trails, data residency controls, and retention policies built into the platform from day one.

### 7) Vendor Independence

No dependency on a single model provider. Cross-Model MoE absorbs the best of the open ecosystem, insulating you from provider pricing changes, policy shifts, or service disruptions.

---

## AVELIN vs Alternatives

| Alternative | How AVELIN Wins |
|-------------|-----------------|
| **Direct frontier providers** | up to 85% cheaper on equivalent quality. Sovereign deployment on your GPUs. No vendor lock-in. |
| **OpenRouter / API aggregators** | Cross-Model MoE is trained AI technology — not pass-through routing. |
| **Open-source self-hosted** | Full platform with trained MoE, governance, and continuous improvement. Zero maintenance overhead. |
| **Building in-house** | 6–12 months and $200K–$500K to replicate. Deploy AVELIN today. |

---

## Stakeholder Value

| Stakeholder | What Improves |
|-------------|--------------|
| **Executive leadership** | Sovereign AI capability on your own infrastructure. Faster decisions with frontier-quality intelligence. |
| **Engineering teams** | Coding models benchmarked at 42–52 on AA Coding Index, at half the cost of alternatives |
| **Operations teams** | Agentic models automate complex multi-step workflows reliably |
| **Security & compliance** | Sovereign deployment on your GPUs, audit-ready governance, data residency controls |
| **Finance & procurement** | Enterprise licensing (fixed annual cost — [request a POC](/poc#request)) or usage-based API ($0.285–$12.50/1M tokens) — up to 85% savings vs frontier |

---

## Deployment Options

| Model | Description |
|-------|-------------|
| **Enterprise License** | Deploy on your own GPUs. Fixed annual licensing — [request a POC](/poc#request). Complete sovereignty. |
| **Cloud API** | Immediate access via API. No infrastructure to manage. |
| **Private cloud** | Dedicated deployment in your cloud tenancy (AWS, Azure, GCP). |
| **On-premises** | Full platform inside your data center. Complete data sovereignty. |
| **Air-gapped** | Operation without external connectivity. For classified and regulated environments. |

All deployment options include the full Cross-Model MoE portfolio, governance controls, and platform capabilities.

---

## Recommended Next Reading

- Functional architecture: [`architecture.md`](architecture.md)
- Systems and product catalog: [`systems/index.md`](systems/index.md)
- Business use cases: [`use-cases/index.md`](use-cases/index.md)
- Benefits and value: [`benefits/index.md`](benefits/index.md)
- Operations and governance: [`operations/index.md`](operations/index.md)
- Business glossary: [`glossary.md`](glossary.md)