A third-party security analysis reported that Cactus Hybrid cloud handoff can disable TLS peer and hostname verification by default unless CACTUS_CLOUD_STRICT_SSL is set, which could expose handoff traffic and tool context.
SourceKernels & Engine for running AI on consumer devices.
Public company, workplace, funding, and market signals
Updated Jul 30, 2026
Cactus Compute is a San Francisco-based YC S25 startup building an open-source, hybrid on-device AI inference engine for phones, laptops, wearables, and other edge devices, with cloud handoff for harder requests.
Primary product
Cactus Engine
Founded
2025
Headquarters
San Francisco, California, United States
Team size
20-30
Industry
Research Services
Sub-industry
on-device AI inference / edge AI infrastructure
Offices
0 jobs at Cactus
Check back later for new openings
Business model
Stage
pre-seed/seed
Total raised
$625K
Latest round
Pre-Seed Round · Oct 2025
Latest amount
$500K
Oct 2025 · Y Combinator
Aug 2025
Investors
Open-source, research-heavy, and performance-focused; the company appears mobile-first, shipping-oriented, and distributed across multiple countries.
Pricing
Freemium / usage-based cloud inference: free on-device tier with paid cloud STT/LLM and hybrid features for production use.
Differentiators
Technology
Customers
Competitors
Estimated monthly visits
19.5K
A third-party security analysis reported that Cactus Hybrid cloud handoff can disable TLS peer and hostname verification by default unless CACTUS_CLOUD_STRICT_SSL is set, which could expose handoff traffic and tool context.
SourceGround Truth · Jul 2026
Third-party security analysis highlighted a cloud-handoff TLS verification concern in Cactus Hybrid and described the model's confidence-based routing design.
LinkedIn · Jul 2026
Major product release introducing a PyTorch transpiler, lossless low-bit quantization, GPU support, and hybrid routing.
LinkedIn · Jun 2026
Cactus said it powers the AnythingLLM mobile app, citing the app's scale and the partnership as a production deployment.
Cactus Blog · May 2026
Official launch of Needle, a 26M parameter function-calling model with high throughput on consumer devices and open-source weights on Hugging Face.
Cactus Blog · Apr 2026
Announced day-one Gemma 4 support, multimodal on-device inference, and cloud handoff for harder tasks.