LLM Credentials

• Keys stay on the server and are injected at session creation
• Per-tenant isolation for multi-tenant deployments

Passing API keys

Pass LLM provider keys via the env option on createSession. The VM does not inherit from the host process.env, so keys must be passed explicitly.

const session = await agent.createSession("pi", {
  env: { ANTHROPIC_API_KEY: process.env.ANTHROPIC_API_KEY! },
});

Per-tenant credentials

Look up each tenant’s API key from your database and pass it at session creation.

import { agentOs } from "rivetkit/agent-os";
import { setup, UserError } from "rivetkit";
import common from "@rivet-dev/agent-os-common";
import pi from "@rivet-dev/agent-os-pi";

interface ConnState {
  userId: string;
  anthropicApiKey: string;
}

const vm = agentOs({
  createConnState: async (c, params: { authToken: string }): Promise<ConnState> => {
    const user = await validateAndLookupUser(params.authToken);
    if (!user) {
      throw new UserError("Forbidden", { code: "forbidden" });
    }
    return { userId: user.id, anthropicApiKey: user.anthropicApiKey };
  },
  options: { software: [common, pi] },
});

export const registry = setup({ use: { vm } });
registry.start();

Then use the connection state when creating sessions:

const session = await agent.createSession("pi", {
  env: { ANTHROPIC_API_KEY: c.conn.state.anthropicApiKey },
});

See Sandbox Agent’s LLM credentials documentation for more details on per-tenant token patterns.

Embedded LLM Gateway

The Embedded LLM Gateway (coming soon) will remove the need to manage API keys manually. It routes all agent LLM requests through a managed proxy built into agentOS, providing per-tenant usage metering, rate limiting, and cost controls without deploying a separate gateway service.