Models & routing
Every step gets
the model it needs.
ORVYN routes each step of a mission to a suitable model lane — or to a model you run yourself. Same tools, same permissions and same verification, whichever model does the thinking.
Routing lanes
Five lanes, chosen per step.
Auto looks at what each step actually requires and picks a lane. You can override it at any time.
Auto
ORION picks per step. Recommended for most missions.
Fast
Low latency and low cost: lookups, summaries, simple edits.
Code
Repository work: multi-file edits, refactors and tests.
Frontend
Vision and layout: reading screenshots, fixing UI at breakpoints.
Premium
Complex, multi-system reasoning where getting it right matters most.
How routing decides
What task analysis looks at.
Routing is a per-step judgment, logged in the mission so you can see why a lane was chosen.
Coding ability
Steps that change code across files go to lanes that are strong at repository work.
Vision
Steps that read screenshots or layouts need a model that can see.
Tool support
Steps that call many tools need reliable tool use through the ToolGateway.
Complexity
Planning across systems gets more reasoning; a rename does not.
Cost
When two lanes can do a step equally well, the cheaper one wins.
Bring your own models
Use the models you already run.
Add a model endpoint in Settings → Models and it appears as a choice next to the built-in lanes.
Local models
Point ORVYN at a model served on your own machine, such as an Ollama server.
Your GPU servers
Use inference servers you host — for example vLLM or llama.cpp — inside your network.
Custom HTTP endpoints
Connect other model APIs your organization has approved.
Same guardrails
Your models call tools through the same gateway, with the same approvals and audit trail.
Available endpoints and providers can depend on your plan and your organization’s policy.
Control and cost
Predictable when you need it.
Routing is the default, not a requirement.
- Pin a model for a workspace or a single mission
- Restrict which models an organization may use
- See tokens used per mission and per cycle in the app
- Set token budgets so a mission stops and asks instead of overspending
- Every model goes through the same verification before a result is reported
Questions
Model and routing questions
Which AI model does ORVYN use?
ORVYN routes each step to a suitable model lane — Fast, Code, Frontend or Premium — based on the step’s needs. You can also pin a specific model for a workspace or mission.
Can I bring my own LLM to ORVYN?
Yes. You can add model endpoints, such as a local Ollama server or an inference server you host, and choose them like any other model. Availability can depend on your plan and organization policy.
Do I need my own API keys?
No. ORVYN routes to models for you. Adding your own endpoints is optional.
Does model choice change what ORION is allowed to do?
No. Every model uses the same ToolGateway with the same permissions, approvals and audit trail.
How does routing reduce cost?
Simple steps go to fast, low-cost lanes and only demanding steps use premium reasoning. The mission summary shows tokens used.
Stop prompting.
Start delegating.
Give ORVYN the goal and stay focused on what matters.