Refiant

What Refiant does

Refiant provides compact, energy-efficient AI models that reason over extensive contexts like whole codebases and long-running agents. These models are delivered with reduced infrastructure, cost, and energy footprints. They compress models and expand context windows to fit on single GPUs and edge devices, maximizing throughput with minimum latency. Refiant offers access to long context models via API (1M, 5M, 10M context windows) or a chat interface for document uploads. Private deployment options are also available to retain data sovereignty. The company's proprietary compression and context-management techniques enable long context without the usual infrastructure tax, offering benefits like lower costs, improved latency, privacy-first deployment, and efficient energy management.

Products

• Long Context Protea Series ModelsEnergy efficient long context models accessible via API or Chat interface, with options for 1M, 5M, and 10M context windows. Also available for private or local deployment.
• Contextual IntelligenceModel inference compression offering contextual intelligence for scale, with 1M+, 5M+, 10M+ context windows.

Key capabilities

• Compact AI models
• Long context windows (1M, 5M, 10M tokens)
• Reduced infrastructure footprint
• Lower cost
• Energy efficient
• Fits on single GPUs and edge devices
• Maximizes throughput
• Minimum latency
• API access
• Chat interface for document upload
• Private deployment for data sovereignty
• Proprietary compression techniques
• Context management techniques
• Low memory footprint
• Low compute footprint

Use cases

• Reasoning over whole codebases
• Long running agents
• Messaging (5 years of email or Slack messages)
• Documentation (10 years of reports)
• Agentic Workflows (24/7/365 agents with real-time data inputs)

AI approach

Refiant develops compact AI models that reason over large contexts, such as whole codebases and long-running agents, with reduced infrastructure, cost, and energy footprints. They achieve this through proprietary compression and context-management techniques, offering models with 1M, 5M, and 10M context windows.

Tech named: Machine Learning, AI, Optimization, long context models, compression, context management, Protea Series Models

Industries served

• Insurers
• Pension Funds
• Data-driven NGO

What it says sets it apart

• Compact AI models that reason over more (whole codebases and long running agents)
• Delivered at a reduced infrastructure, cost and energy footprint
• Compresses models and expands context windows to fit on single GPUs and edge devices
• Maximises throughput with minimum latency
• Offers long context models via API or Chat interface
• Provides private deployment for data sovereignty
• Combines compression and context management to focus only on what matters efficiently
• Little to no loss of reasoning quality relative to heavier models
• Long-context without the usual infrastructure tax
• Built on proprietary compression and context-management techniques
• Addresses the limitations of models with limited context, avoiding RAG, chunking, and summarization workarounds

Funding rounds we track

From the AI funding tracker — rounds as reported by the linked publications.