Subconscious

The product is an inference system and runtime for long-horizon coding agents and agentic products that addresses context limits, sustained throughput, and cost for workloads spanning millions of tokens and hundreds of steps. It sells to developers and engineering teams building coding agents and agentic products, including enterprises that need dedicated capacity for internal agent fleets. Compared to standard inference, it is positioned as an optimized runtime with post-trained open models that runs via a hosted API or self-hosted in the customer's cloud.

Key features

• 5m+ token context window
• 3.5x faster sustained throughput
• 40% savings on inference costs
• Runtime context compression
• 2.3x concurrent workloads on same hardware
• OpenAI and Anthropic compatible API
• Self-host in your cloud
• Dedicated GPU infrastructure with no rate limits
• Post-training on tools and data
• Open post-trained Marathon models
• Support for GLM, Qwen, Nemotron, Gemma, DeepSeek
• Plug into Claude Code, Codex, OpenCode harnesses