RunInfra
Hosted open-model inference API for AI agents with session-aware cache routing, compatible with the OpenAI and Anthropic SDKs.
About RunInfra
RunInfra, made by RightNow AI, Inc., is a hosted Model API service that serves open models to AI agents and coding tools through a single API key. Chat completions and Responses endpoints are OpenAI-compatible, and a Messages endpoint is Anthropic-compatible. The homepage lists models such as Ornith, DeepSeek, GLM, Qwen and Nemotron variants.
What It Is
RunInfra is an inference gateway for open models. Developers set the base URL to the RunInfra API, use a workspace API key, and pass a model id, so existing OpenAI or Anthropic SDK code can call hosted models. The model library shows per-model input, cached input and output prices, output speed, and capabilities such as tool calling, JSON mode, streaming and image input.
How Cache-Aware Routing Works
RunInfra describes session-aware routing that sends repeat calls from one agent session to the replica already holding its cached context. Routing is automatic from the conversation or workspace, or callers can send a session ID to keep calls together. Cache hits are reported in usage and billed at the cached rate, and the site notes hits are best effort. The site publishes a cache measurement method and shows effective input rates based on measured cache share over the last 24 hours.
Coding Agent Setup
A CLI and a copy-paste setup prompt let coding agents such as Claude Code, Codex, OpenCode, Cline, Droid and Grok Build configure RunInfra models. The agent reads the setup docs, the user approves sign-in in the browser, and the user confirms a setup plan before anything is installed or changed. Docs also cover OpenAI compatibility with LangChain, LiteLLM, LlamaIndex and the AI SDK, rate limits and troubleshooting.
Data Handling and Security
RunInfra states that hosted Model APIs never store prompts, with zero data retention by default and no training on prompts or completions. It lists TLS in transit, AES-256 at rest, workspace roles with row-level security, and a SOC 2 Type II attestation with the report available under NDA. The site also references an enterprise serving package license for deploying optimized serving packages in a licensee's own products.
Community Discussions
Be the first to start a conversation about RunInfra
Share your experience with RunInfra, ask questions, or help others learn from your insights.
Pricing
Pay as you go
Hosted Model APIs billed per token from a prepaid balance; no subscription. $5 minimum top-up; credits last a year.
- $5 minimum top-up
- OpenAI-compatible chat completions and Anthropic-compatible Messages
- Cached input discounts on supported models
- Ornith 1.5 122B: input $0.26, cached input $0.07, output $2.08 per 1M tokens
- DeepSeek V4.1 Flash 1017: input $0.15, cached input $0.01, output $0.60 per 1M tokens
- GLM 5.3: input $1.30, cached input $0.13, output $4.20 per 1M tokens
- GLM 5.3 Flash: input $0.11, cached input $0.03, output $0.45 per 1M tokens
- Nemotron 3.5 Lightning 30B: input $0.05, cached input $0.01, output $0.15 per 1M tokens
- Get $1 free to test the Model APIs
Starter
Coding plan with about $12 of usage a month
- About $12 of usage a month
- 5-hour limit $1.11
- Weekly limit $2.77
- Free Standby up to $0.20 of chat a day
- Requests at once per model, when busy 4 to 96
Pro
Coding plan with about $35 of usage a month
- About $35 of usage a month
- 5-hour limit $3.23
- Weekly limit $8.08
- Free Standby up to $0.58 of chat a day
- Requests at once per model, when busy 5 to 96
Team
Coding plan with about $119 of usage a month
- About $119 of usage a month
- 5-hour limit $10.98
- Weekly limit $27.46
- Free Standby up to $1.98 of chat a day
- Requests at once per model, when busy 8 to 96
Enterprise
Dedicated infrastructure, compliance, and custom volume.
- Audit logs and role-based access control (RBAC)
- Custom volume and contract terms
- Custom SLAs up to 99.99%
- SOC 2 Type II attestation
- Dedicated CSM and private Slack
Capabilities
Key Features
- OpenAI-compatible chat completions and Responses endpoints
- Anthropic-compatible Messages endpoint
- Session-aware routing to warm cached replicas
- Cached input billing with reported cache hits
- Hosted open model library with tool calling, JSON mode, streaming and image input
- CLI and prompt-based setup for coding agents
- Zero data retention by default for hosted Model APIs
- Workspace API keys and roles
- Published cache measurement methodology