Where is OpenSquilla headquartered?

OpenSquilla is headquartered in Shanghai, China. The company has a team of 101-500 people.

What does OpenSquilla do?

OpenSquilla operates in the AI Agents sector. Token-efficient, open-source microkernel AI agent runtime with intelligent model routing, multi-tier memory, and secure sandboxing, developed by TokenRhythm Technologies (基元律动). Token-efficient AI agent that delivers higher intelligence density for the same budget — 60-80% token cost reduction through intelligent model routing, multi-tier memory, and on-device ML optimization.

When was OpenSquilla founded?

OpenSquilla was founded in 2026. It is based in Shanghai, China.

Who is the CEO of OpenSquilla?

Yunhe Wang (王云鹤) is the CEO of OpenSquilla. The company was founded in 2026.

Where is OpenSquilla located?

OpenSquilla is headquartered in Shanghai, China.

What funding stage is OpenSquilla at?

OpenSquilla is currently at the Early Stage stage.

How many employees does OpenSquilla have?

OpenSquilla has approximately 101-500 employees.

OpenSquilla

Category: AI Agents

Token-efficient, open-source microkernel AI agent runtime with intelligent model routing, multi-tier memory, and secure sandboxing, developed by TokenRhythm Technologies (基元律动). OpenSquilla was founded in 2026. The company is led by Yunhe Wang (王云鹤). Based in Shanghai, China. Team size: 101-500. Latest round: Angel. Key investors include Top-tier Chinese VC firms, Major Chinese internet companies (specific names not publicly disclosed).

Founded: 2026
Headquarters: Shanghai, China
Team size: 101-500

Value proposition

Token-efficient AI agent that delivers higher intelligence density for the same budget — 60-80% token cost reduction through intelligent model routing, multi-tier memory, and on-device ML optimization.

Products and solutions

OpenSquilla AI Agent Runtime (open-source, Apache 2.0), SquillaRouter (on-device ML model router), Multi-Model Ensemble Routing (v0.5.0), MetaSkill self-organizing skill protocol, Desktop apps (macOS/Windows), CLI and Web UI, Multi-channel support (Slack, Discord, Telegram, WeCom, DingTalk

Unique value

Microkernel AI agent runtime that routes each task to the cheapest capable model via on-device ML classifier, achieving 60-80% token cost savings while matching or exceeding top-tier single-model performance through multi-model ensemble routing.

Target customer

Enterprises and developers seeking cost-efficient, self-hostable AI agent deployments for production workloads; Chinese enterprises with compliance requirements; developers building multi-channel AI assistants

Industries served

Enterprise automation, Knowledge management, Customer service, R&D assistance, Office productivity, Government/state-owned enterprise digital transformation

Technology advantage

On-device ML-based intelligent model routing (SquillaRouter) using ONNX Runtime + LightGBM; Multi-model ensemble routing (v0.5.0) that combines 4 Chinese foundation models to beat Anthropic Fable 5 at 1/3 cost; 4-tier cognitive memory architecture (working, episodic, semantic, raw); Syscall-level sandbox isolation; Apache 2.0 open-source; Supports 20+ LLM providers

How they differentiate

Unlike single-model agent frameworks (OpenClaw, AutoGPT, CrewAI), OpenSquilla uses a microkernel architecture with ML-driven routing that dynamically assigns tasks to the optimal model. Its v0.5.0 multi-model ensemble routing combines multiple Chinese foundation models to outperform top-tier proprietary models at a fraction of the cost. The company is founded by ex-Huawei Pangu model lead with deep expertise in model compression and efficient inference.

Main competitors

OpenClaw (open-source AI agent framework), AutoGPT, CrewAI, LangGraph

Key partnerships

Infinigence AI, City University of Hong Kong, SEE Fund, Peking University, Shanghai Jiaotong University, Beijing Jiaotong University (joint research on Claw-SWE-Bench benchmark)

Notable customers

State-owned enterprise clients (国资背景大厂客户) — specific names not publicly disclosed

Major milestones

2026-04-08: Company registered (Shanghai TokenRhythm Technology Co., Ltd.), 2026-05: v0.1.0 launched (smart routing), 2026-05: v0.2.0 (1-click migration), 2026-06: v0.3.0 (MetaSkill protocol), 2026-06: Angel round at $100M valuation, 2026-06: v0.4.0 (coding mode, desktop apps), 2026-07: v0.5.0 Preview (multi-model ensemble routing tops DRACO benchmark), 2026-07: Agentic Routing technical report published, 2026-07-05: v0.5.0 Preview 2 released, 2026-07-06: Usage guide published

Growth metrics

GitHub: 5,400+ stars, 383 forks, 834 commits, 10+ contributors (as of July 2026); X/Twitter: 6,667 followers

Market positioning

Emerging Chinese AI agent platform positioned at the intersection of cost-efficiency and enterprise readiness, competing with open-source agent frameworks while leveraging unique model routing IP from ex-Huawei Pangu team.

Geographic focus

China (primary); Global open-source community

Patents and IP

GhostNet neural architecture (co-developed at Huawei); "High-efficiency multiplier and adder neural network" — Huawei Top 10 Invention; Agentic Routing harness-native data flywheel methodology

About Yunhe Wang (王云鹤)

Ex-Huawei Noah's Ark Lab Director & Pangu Large Model Lead; PhD in Intelligent Science from Peking University (2018); BSc in Mathematics from Xidian University; Google Scholar citations 30,000+; Co-creator of GhostNet neural architecture; Huawei "Top 10 Invention" awardee

Official website: https://opensquilla.ai