-
orcarouter/Nex-N2.5-mini-Uncensored
Text Generation • 35B • Updated • 962 • 10 -
orcarouter/Nex-N2.5-mini-Uncensored-FP8
Image-Text-to-Text • 35B • Updated • 201 • 5 -
orcarouter/Nex-N2.5-mini-Uncensored-NVFP4
Image-Text-to-Text • 35B • Updated • 249 • 6 -
orcarouter/Nex-N2.5-mini-Uncensored-MLX
Image-Text-to-Text • Updated • 10
AI & ML interests
The AI Gateway built for Privacy & Security🐳 200+ models. 0% markup. BYOK. Adaptive Routing. Firewall. Guardrails. Observability. Cyber. Join the builders→ https://discord.gg/YEubt8enRA
Recent Activity
OrcaRouter
Open AI infra, models, and security research.
Continuum AI builds secure and accessible infrastructure for AI inference, model routing, agent systems, model optimization & quantization, and cyber. We maintain OrcaRouter, a privacy- and security-first AI gateway for open and frontier models, alongside open-source research and tooling for adaptive routing, agent observability, cybersecurity, red teaming, model evaluation, and efficient local inference.
Our goal: make intelligence accessible to everyone — through one gateway, on any hardware, without compromising security or privacy.
Our Hugging Face releases include production models, quantized models, and research-oriented variants for authorized security research, adversarial evaluation, defensive cybersecurity, and model behavior research.
Open Source
- OrcaRouter Lite — self-hosted AI router
- OrcaReplay — record, replay, and fork AI agent runs
- OrcaPromptVault — archive of agent system prompts and tool schemas
- OrcaCode Review — AI code review and merge gates
- Orca AI Incident Archive — real-world AI agent security incidents
- SCUTTLE — hybrid post-quantum encryption for data
- OrcaSAQ2 Kernel — serving stack for compressed models
Selected Research
- A Production-Oriented LLM Router with Hybrid Offline-Online Learning
Contextual-bandit routing with hybrid offline-online learning for production AI infrastructure. - How Fragile Is Safety Alignment at Frontier Scale?
A single-direction white-box attack on safety alignment in a 320B mixture-of-experts model. - Better Retrieval, Worse Robustness — EMNLP 2026 Main Conference
How multi-hop RAG can amplify upstream speech-recognition errors. - IndexRAG — AACL-IJCNLP 2026 Findings
Index-time bridging facts for efficient cross-document and multi-hop reasoning.
🌐 https://www.orcarouter.ai
🤖 https://www.orcarouter.ai/models
💻 https://github.com/Continuum-AI-Corp
-
orcarouter/Nex-N2.5-mini-Uncensored
Text Generation • 35B • Updated • 962 • 10 -
orcarouter/Nex-N2.5-mini-Uncensored-FP8
Image-Text-to-Text • 35B • Updated • 201 • 5 -
orcarouter/Nex-N2.5-mini-Uncensored-NVFP4
Image-Text-to-Text • 35B • Updated • 249 • 6 -
orcarouter/Nex-N2.5-mini-Uncensored-MLX
Image-Text-to-Text • Updated • 10