Features
Reconfigurable dataflow architecture (XPU)
Supports FP8 to FP64 precision
Dense and irregular workload optimization
Minimizes data movement for energy efficiency
FPGA prototype (Grasshopper) for early testing
Cluster design (Monolith) for scaling as a single machine
Aimed at AI discovery loops (propose, simulate, test, learn)
Designed for training, simulation, and verification
Low-energy AI inference and training
Hardware for recursive self-improvement and superintelligence
Real-time web search API for agents
Content extraction and cleaning
/research endpoint with state-of-the-art benchmarks
Dynamic filtering — model programs own search filters via Bash/Python
180 ms p50 median latency on /search
Intelligent caching and indexing
300M+ monthly request capacity
Built-in PII, prompt injection, and malicious source filters
Drop-in integration with OpenAI, Anthropic, Groq
x402: pay-per-query with USDC wallet on Base
Integration with Arcade.dev, LangChain, MCP Marketplace
Structured, chunked output for LLM ingestion
Developer-friendly API with comprehensive docs
