RunInfra
RunInfra provides session-aware routing for AI inference that keeps cached context warm across repeated calls, reducing latency and cost for token-heavy agent workloads. Its routing layer pins each conversation to a single replica, improving cache hit rates, and exposes a unified API compatible with OpenAI and Anthropic SDKs. The service includes a CLI that automates setup across 17+ coding agents including Claude Code, Codex, and Cursor.
- Artificial Intelligence
- Developer Tools
- Software Only
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
AI agent workloads generate highly repetitive call patterns, where the same conversation context is sent repeatedly to an inference provider. Without session awareness, each repeat request can land on a different replica, missing the cached prefix and forcing full recomputation—raising both latency and cost for token-heavy agent sessions.
Solution
RunInfra provides session-aware routing for AI inference that keeps cached context warm across repeated calls. Its routing layer learns which replica holds a given conversation's context and directs repeat calls to that same replica, improving cache hit rates dramatically. The service exposes a unified API compatible with OpenAI and Anthropic SDKs, so developers can switch without changing their stack. Cache hits are reported in usage data and billed at a discounted cached rate, while effective input prices are published transparently per model. A model library spans open-weight models like GLM 5.3 Flash with measured speed, quality, and cache-hit statistics. A CLI and automated setup flow let coding agents configure the service themselves through a browser-based approval process.
Target Audience
Primary customers are development teams running AI coding agents, agentic workflows, and other multi-turn inference workloads that need low latency and predictable costs. The service also targets enterprises integrating open-weight models into production applications through standard SDKs.
Features
- Session-aware routing that pins each conversation to a single replica, avoiding cold-cache recomputation on repeat calls
- Unified API compatible with OpenAI and Anthropic SDKs, requiring only a base URL change for most integrations
- Transparent usage reporting that distinguishes cached vs. uncached tokens, with cache hits billed at a discounted per-token rate
- Published measurement methodology covering throughput, time-to-first-token, latency percentiles, and quantization quality recovery
- CLI supporting automated setup across 17+ coding agents including Claude Code, Codex, OpenCode, Pi, Aider, Qwen Code, and Cursor
- Effective input price calculations based on measured cache hit rates on real billed traffic, with methodology disclosed publicly