Introducing

EdgeRunner Athena

EdgeRunner Athena is your air-gapped personalized AI assistant for improved productivity, decision making, and operational efficiency.

Powered by EdgeRunner’s own language models, Athena is the ultimate compression function for knowledge at the edge. Athena is completely air-gapped, ensuring the highest level of privacy and data protection. EdgeRunner Athena provides a ChatGPT-like experience without requiring internet, running 100% locally and privately, powered by EdgeRunner-Tactical-7B, our latest SOTA model, competitive with Meta’s Llama-3-70B.

KEY BENEFITS

Ownership

Ownership

Full control over your AI—no more renting a black box.

Ownership

Ownership

Full control over your AI—no more renting a black box.

Privacy

Privacy

Your data stays on your device, secure and private. Never send data to a 3rd party again.

Privacy

Privacy

Your data stays on your device, secure and private. Never send data to a 3rd party again.

Personalization

Personalization

Trained on your unique data, providing answers that fit your specific needs.

Personalization

Personalization

Trained on your unique data, providing answers that fit your specific needs.

Transparency

Transparency

Built on open-source models. You have full visibility into how it works and how it’s improved.

Transparency

Transparency

Built on open-source models. You have full visibility into how it works and how it’s improved.

Productivity

Productivity

Empower your team with AI that enhances productivity and effectiveness.

Productivity

Productivity

Empower your team with AI that enhances productivity and effectiveness.

Custom Deployments

Custom Deployments

More SOTA models available for more basic to more powerful setups.

Custom Deployments

Custom Deployments

More SOTA models available for more basic to more powerful setups.

Infinite Savings

Infinite Savings

Zero inference costs. No token fees—use your AI as much as you need without extra charges.

Infinite Savings

Infinite Savings

Zero inference costs. No token fees—use your AI as much as you need without extra charges.

Power Efficient

Power Efficient

Only requires 4GB of RAM—leverages GPU/NPU but can fall back on CPU only.

Power Efficient

Power Efficient

Only requires 4GB of RAM—leverages GPU/NPU but can fall back on CPU only.

Runs Anywhere

Runs Anywhere

No internet required, running 100% locally and privately.

Runs Anywhere

Runs Anywhere

No internet required, running 100% locally and privately.

KEY FEATURES

Specialized AI Agents

Powered by the EdgeRunner Command function calling model - enabling task execution within agentic workflows.

Our function calling model can automate tasks such as opening Slack, managing emails, browsing, invoking other models, handling Excel, swapping LoRAs, and more.

Air-Gapped Offline Functionality

Operates independently of the internet, safeguarding sensitive data from cyber threats and maintaining operational security.

air-gapped

Hardware & OS Agnostic

Leverages GPU/NPU when available but can fall back on CPU only - only requires 4GB of RAM. Works on any OS.

hardware-agnostic

Natural flow

of communication

Listens to and transcribes meetings and interactions.

image_/images/product/Transcription.webp

Localized RAG (Retrieval-Augmented Generation)

Provides real-time, context-aware recommendations to enhance decision-making and operational efficiency.

rag
Local Data Integration

Local Data Integration

Delivers precise outputs with citations derived from local data sets provided by the user.

Domain-Specific Expertise

Domain-Specific Expertise

Customizable to meet the specific needs of various roles.

Customizable Personas

Customizable Personas

Can be tailored to assist with specific missions and tasks.

Custom Adaptors (LoRA)

Custom Adaptors (LoRA)

Utilizes Low Rank Adaptation of LLMs for task-specific enhancements, tailoring AI capabilities to distinct roles.

Dynamically switching between SLMs.
  

Intelligent routing of requests: EdgeRunner Athena will intelligently switch between models, dynamically routing requests to the best task-specific model for the use case.

Power efficiency: Saves RAM and power, increasing efficiency and performance.

Streamlined accessibility: Makes Generative AI simple and boring, enabling widespread adoption.

Ever-evolving standard: Becomes the enterprise standard for leveraging multiple models at once, continuously leveraging the latest models, future proofing intelligence.

MODELS

Small, open models are the solution.
Swarm intelligence

Swarm intelligence

Multiple tiny, task-specific models working in unison provide better results than large general purpose models.

Swarm intelligence

Swarm intelligence

Multiple tiny, task-specific models working in unison provide better results than large general purpose models.

Higher performance

Higher performance

Large general purpose models (e.g. Llama 3) are deeply degraded after 2-Bit quantization and distillation to fit on chip.

Higher performance

Higher performance

Large general purpose models (e.g. Llama 3) are deeply degraded after 2-Bit quantization and distillation to fit on chip.

Task-specific

Task-specific

To solve real challenges, enterprises need task-specific models running locally rather than general purpose models.

Task-specific

Task-specific

To solve real challenges, enterprises need task-specific models running locally rather than general purpose models.

Offline computing

Offline computing

Due to data gravity, AI will increasingly move to the edge, which requires smaller models that can run without connection to the internet.

Offline computing

Offline computing

Due to data gravity, AI will increasingly move to the edge, which requires smaller models that can run without connection to the internet.

Data transparency

Data transparency

Enterprises and governments will require open models with open datasets due to concerns around IP, explainability, and bias.

Data transparency

Data transparency

Enterprises and governments will require open models with open datasets due to concerns around IP, explainability, and bias.

TACTICALTACTICAL

INTRODUCING

EdgeRunner-Tactical-7B

EdgeRunner Tactical-7B

Context length of 128K tokens

A top performing 7B parameter in the world

Context length of 128K tokens

A top performing 7B parameter in the world

Initialized from Qwen2-Instruct, leveraging prior advancements.

Competitive performance with Llama-3-70B, Mixtral 8x7B, and Yi 34B.

More performant than models as large as 30B.

COMMANDCOMMAND

INTRODUCING

EdgeRunner-Command-7B

EdgeRunner Command-7B

Designed specifically for Function Calling

A SOTA 7B parameter language model

Designed specifically for Function Calling

A SOTA 7B parameter language model

Advanced Function Calling for Air-Gapped Workflow Execution: Excels at interpreting, executing, and chaining function.

Dual-Mode Functionality: Can serve as a tool router for request analysis and routing & standalone Chat Agent.

Guaranteed Helpful Ability: Achieved strong scores on popular benchmarks, including Arena Hard Benchmark and MT-Bench.

CASE STUDY

How Cachengo became AI-enabled

Cachengo is the industry leader in incorporating machine learning (ML) capabilities into smart appliances to not only enable dynamic policies and artificial intelligence, but to also dramatically cut power consumption. With an emphasis on 100% open source and reference hardware designs, Cachengo is changing the landscape for how we envision servers and storage at the Cloud Edge and for 5G. EdgeRunner AI is helping Cachengo transform their "content addressable storage" into "context addressable storage" using AI at the edge.