01
ProfileCTO / Head of Platform
A single point of access to AI, with no vendor lock-in, installed in your environment and with aliases that decouple the application from the model.
Primora Orbit · AI Governance
All of your company’s AI, in one place and under control.
Primora Orbit centralizes access to your company’s AI models in a single point, installed in your environment, auditable and compatible with the OpenAI API.
Teams pick the right model for each job; the company decides who uses what, tracks consumption and keeps data inside its own perimeter.
Illustrative values
Compatible with OpenAI-standard API providers and the leading coding assistants.
OpenAIAzure OpenAIMicrosoft FoundryOpenRouterMistralGroqOllamavLLMLiteLLMClaude CodeCodexOpenCodeHermes
The problem
Your team already uses AI. Who is in control?
Without a central point, each area adopts AI on its own. And the risks pile up without anyone noticing.
API keys scattered across repositories, spreadsheets and chats, with no owner and no expiry.
01Coding assistants talking directly to providers, like Claude Code and Codex, with no control point in the path.
02Sensitive data leaving the company in prompts nobody reviewed: IDs, contracts, source code.
03Lock-in to a single vendor switching models means touching every application.
04No audit trail to answer legal, the DPO or the auditor.
05
The solution
One entrance for AI, with rules at every step.
Primora Orbit sits between applications and providers. Every request follows the same route, checked and recorded:
For builders
Your code stays the same.
Same OpenAI-standard API; only the address and the credential change.
from openai import OpenAI
client = OpenAI(
base_url="https://orbit.yourcompany.com/v1/inference",
api_key="ock_••••••••", # only the URL and the credential change
)
resp = client.chat.completions.create(
model="chat-default",
messages=[{"role": "user", "content": "..."}],
)For governance
Every request leaves a trail.
Model, applied rule and consumption are recorded for review.
- runtime
- claude-code · platform
- alias
- chat-default → approved model resolved
- policy
- access by credential applied
- limit
- 4,200 / 5,000 requests 84%
- content
- prompt and response not stored
- audit
- event #18,204 chained
Who it is for
Built for those who decide, protect and build.
Technology, security, legal and business look at AI differently. Orbit serves each of them.
02
ProfileCISO / Security
Deny-by-default model access, write-only credentials, content protection and a hash-chained audit trail.
03
ProfileDPO / Legal
Prompts and responses never stored, no telemetry by default and data inside your infrastructure.
04
ProfileManagement / FinOps
Usage limits on requests and tokens, with consumption consolidated per credential and per application.
05
ProfileDevelopers
OpenAI-compatible API and assistants like Claude Code, Codex and OpenCode running with orbit run, without changing the workflow.
06
ProfileBusiness teams
Access only to the models approved by the company, with corporate login and conversations that are not stored.
How it works
From installation to daily use in five stages.
Demo · Providers and credentials1/5
Your models, connected.
- OpenAI online
- Azure OpenAI online
- Ollama online
Protected credentialwrite-only · 0600 file
Illustrative values
Why Primora Orbit
Control for the company, freedom for the teams.
Several providers and assistants orbiting the same control point.
What makes a difference day to day
Use cases
Orbit in practice.
Real situations where the same governance layer serves very different areas.
Software development
Teams use Claude Code, Codex and OpenCode with governed credentials, approved models per project and recorded usage.
Customer service
Assistants that query the knowledge base, with content protection before the request reaches the model.
Legal and compliance
Document analysis without prompts and responses being stored, with an audit of every decision.
Operations agents
Agents that use tools via MCP, with the tool inventory and classification done by the administrator.
Content that cannot leave
Aliases pointing to models running in your environment, such as Ollama, for cases where the content cannot leave the house, with the same rules and the same audit.
Frequently asked questions
What people usually ask us.
Did not find what you were looking for? Our team explains how Orbit fits your scenario.
Talk to a specialistNo. If the application already uses an OpenAI-compatible SDK, just point the base URL to Orbit and use a runtime credential.
Any provider with an OpenAI-compatible endpoint, such as OpenAI, Azure OpenAI / Microsoft Foundry, OpenRouter, Mistral, Groq and Ollama. With the optional LiteLLM-based gateway, the list grows. Every model sits behind an alias.
In your environment, with Docker Compose on Linux. The minimal install uses about 41 MiB of memory at rest and does not depend on any external service to run.
No. Prompts and responses are transient and are never written to logs, audit or database. Orbit records only metadata: who asked, which model, which decision and how much was used.
With limits on requests and tokens, applied per credential and per application, and with consumption consolidated in the admin console.
Yes. Orbit integrates the Claude Code, Codex, OpenCode and Hermes coding assistants, which then use only the models and rules defined by the company.
Not necessarily. Besides local login, Orbit supports SSO via OIDC, such as Microsoft Entra ID.
We support the rollout alongside your team: assessment, installation, integration with your providers and systems, policy definition and ongoing operation.
Take your company’s AI to production with governance from the start.
Book a conversation with our team and see Primora Orbit applied to your models and your policies.
Or write to comercial@primora.com.br