Use your AI gateway
Route agent inference through the gateway your organization already governs.
"The gateway" in Proliferate means one of two things, and it's worth being clear on which one you're choosing per agent.
Option 1: Proliferate's built-in gateway
Proliferate runs its own model gateway (built on LiteLLM) that any gateway-capable agent (Claude Code, Codex, OpenCode, Grok) can use with the Proliferate gateway switch in that agent's Authentication settings. No key to create or rotate.
Under the hood, the gateway owns:
- Central provider keys, so no one on your team needs their own.
- A per-user virtual key, so usage is attributable to the person who spent it even inside a shared organization.
- A budget enforced at the gateway itself: every account gets a one-time free credit at signup, organizations on a paid plan get a shared team budget, and calls fail closed once the budget is spent.
- Optional auto top-up so the team budget doesn't run out mid-sprint.
Settings > Agents > Codex > Authentication, gateway enabled
The Authentication section with the Proliferate gateway switch turned on and its subtitle reading 'Proliferate-managed model access. No setup required.', with the API key rows below it disabled or hidden.
Option 2: Point an agent at your own gateway
If your organization already runs its own LLM gateway (an internal LiteLLM
deployment, Bedrock, Azure OpenAI, Portkey, or similar), you don't have to
route through Proliferate's. Instead, in that agent's Authentication
section, add a variable for the base URL your gateway expects (for example
ANTHROPIC_BASE_URL or OPENAI_BASE_URL) alongside the key or token your
gateway issues, and leave the Proliferate gateway switch off. The agent talks
straight to your gateway from then on.
Settings > Agents > Claude > Authentication, custom base URL
An API key row with the env var field edited to a custom name like ANTHROPIC_BASE_URL, paired with a second row for the matching key, both enabled, and the Proliferate gateway switch off.
Whichever gateway you use, the model traffic never enters your Proliferate account's audit trail beyond what the gateway itself reports back, and raw provider secrets are never written into the sandbox when you use Proliferate's own gateway (only a scoped virtual key is).
Your own gateway can own whatever it already owns: provider keys, model routing, rate limits, spend controls, audit logs, and data handling rules. Proliferate runs the workspace, agent lifecycle, tools, and review flow on top of it either way.