OrcaRouter
An OpenAI-compatible AI gateway that routes prompts across 200+ models with guardrails, an agent firewall, observability and zero token markup.
At a Glance
About OrcaRouter
OrcaRouter is an AI gateway that gives developers one OpenAI-compatible endpoint to more than 200 models. It grades each prompt, routes it to a suitable model, and layers on guardrails, an agent firewall, prompt caching and per-request logging. The vendor states it adds $0 per token and that prompts are not logged unless logging is turned on.
What It Is
OrcaRouter sits between an application and many model providers. Teams change the SDK base_url to the OrcaRouter API and swap their key; the vendor says existing OpenAI, Anthropic or Google SDK code keeps working for Chat Completions, Responses, Embeddings, Images, Audio and streaming. Requests can use orcarouter/auto so the gateway grades the prompt and picks a model, or users can bring their own provider keys.
Routing and Reliability
Adaptive routing learns the quality and cost trade-off from a workspace's own traffic, and routing rules can be written as code using YAML and CEL. Automatic failover retries against fallback capacity before the response starts. Other documented options include Bring Your Own Key, Orca Connect for using your own GPU behind the same key, and OrcaRouter Lite, described as an MIT-licensed gateway for your own infrastructure.
Security, Governance and Observability
Guardrails block PII, secrets, jailbreaks and prompt injection before a model sees them. The Agent Firewall grades each tool and MCP call as allowed, sent for human review, or blocked. Compliance packs map policies to 32 frameworks, and governance features include scoped API keys, budgets, RBAC, audit trails and declared data regions. Request logs show grade, model, latency and cost per call, and OrcaReplay records and replays agent runs.
Cyber Models and Developer Tools
Verified businesses can access OrcaCyber models for authorized security work, including OrcaCyber Zero and OrcaCyber Ultra. Developers also get a browser playground, versioned prompt management, OrcaCode Review for pull requests, and integrations with tools such as Cursor, LangChain, n8n, the Vercel AI SDK and an MCP server.
Community Discussions
Be the first to start a conversation about OrcaRouter
Share your experience with OrcaRouter, ask questions, or help others learn from your insights.
Pricing
Hacker
Free forever, zero markup on all tokens; full gateway with 200+ models and auto-failover.
- Route โ 200+ models, auto-failover
- Observe โ basic dashboard
- Manage โ prompt versioning
- 10 API keys ยท 0% token markup
- No card needed
Team
Still zero markup; pay for team features.
- Everything in Hacker
- Up to 10 team seats
- Compliance enforcement & reports
- Unlimited API keys
- Priority support
Enterprise
SLA commitments and dedicated capacity with custom pricing.
- Everything in Team
- Unlimited team seats
- Priority access to new models
- Every hidden & beta feature
- Dedicated infrastructure
- 99.99% uptime SLA
- Dedicated support & custom pricing
Capabilities
Key Features
- OpenAI-compatible endpoint for 200+ models
- Adaptive prompt grading and routing
- Automatic failover with fallback chains
- Routing rules DSL (YAML + CEL)
- Bring Your Own Key
- Guardrails for PII, secrets, jailbreaks and prompt injection
- Agent Firewall for tool and MCP calls
- Compliance packs for 32 frameworks
- Request logs and insights
- Prompt caching
- Versioned prompt management
- OrcaReplay agent run replay
- OrcaCode Review for pull requests
- Browser playground for chat, image and video
- Budgets, roles and audit trail
