Migration

Eridian is designed as a drop-in orchestration layer. Most migrations require changing the base URL and API key - the request schema remains OpenAI-compatible. You gain semantic caching, multi-model routing, PII redaction, and cost attribution without rewriting application logic.

From OpenAI Direct

// Before
import OpenAI from "openai";
const client = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });

// After
const client = new OpenAI({
  apiKey: process.env.ERIDIAN_API_KEY,
  baseURL: "https://api.geteridian.com/v1",
});

const response = await client.chat.completions.create({
  model: "auto",
  messages: [{ role: "user", content: "Hello" }],
  // @ts-expect-error Eridian extension
  project_id: "prj_dev_001",
  // @ts-expect-error Eridian extension
  features: ["semantic_cache", "pii_redaction"],
});

Benefits without code restructure: semantic caching, multi-model routing, PII redaction, and unified cost attribution.

From Gemini Direct

# Before
from google import genai
client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])

# After
from eridian import Eridian
client = Eridian(api_key="eridian_sk_...", project_id="prj_dev_001")
response = client.inference.create(
    model="gemini",
    messages=[{"role": "user", "content": "Summarize this filing."}],
)

Set model to "gemini" only when the workflow must stay on Gemini. Prefer "auto" so routing policy can fail over to GPT if Gemini is unhealthy.

From AWS Bedrock

Replace Bedrock model IDs with "auto" or specific Eridian model IDs. Map IAM-authenticated Bedrock calls to Eridian API key authentication:

1. Create an Eridian project and API key with inference:write.
2. Update your application's inference client to target https://api.geteridian.com/v1.
3. Map Bedrock modelId values to Eridian catalog IDs via Models.
4. Enable features incrementally: caching → PII → RAG → routing policies.

Recommended Rollout

Validate cost attribution in the observability dashboard before revoking direct provider credentials.

What Does Not Migrate

Migration does not require moving training data or fine-tuned weights. Eridian routes to frontier models via the orchestration layer. Custom fine-tunes remain on provider accounts until explicitly registered in Eridian model pins.

Rollback Plan

Keep direct provider keys active during canary. Eridian responses include eridian.route so you can compare model behavior. If rollback is required, revert the base URL - no schema migration needed.

See Quickstart for first-call setup, SDKs for Python, TypeScript, and Go, and Governance for budget configuration during rollout.