From Portkey

This guide covers moving a Portkey deployment to HiWay2LLM. Both are OpenAI-compatible routing layers, so the code diff is minimal. The real decisions are around Portkey-specific concepts (virtual keys, guardrails, prompt library) — some map 1-to-1, some don't exist on HiWay. Read the gotchas section before you flip the switch. End-to-end: 10-15 minutes including the provider-keys re-entry.

Step 1: Get your HiWay API key

Open Dashboard → Keys and click New key. Your hw_live_ key replaces the Portkey x-portkey-api-key.

Step 2: Swap your configuration

On Portkey you used the https://api.portkey.ai/v1 base URL with an x-portkey-api-key header and usually a x-portkey-virtual-key header pointing to a stored provider credential. On HiWay you use https://app.hiway2llm.com/v1 with a standard Authorization: Bearer hw_live_... header — the upstream provider is resolved from your enabled-provider list in the dashboard, no extra headers.

from openai import OpenAI

client = OpenAI(
    base_url="https://app.hiway2llm.com/v1",
    api_key="hw_live_YOUR_KEY",
)

response = client.chat.completions.create(
    model="auto",  # or "openai/gpt-4o", "anthropic/claude-sonnet-4-5", ...
    messages=[{"role": "user", "content": "Draft a follow-up email"}],
)

print(response.choices[0].message.content)

Step 3: Verify

Send one request, confirm X-HiWay-Routed-Model on the response, and check Dashboard → Usage for the timeline. If you relied on Portkey's traces view, the HiWay equivalent is Dashboard → Analytics plus the _hiway metadata object on each response.

Gotchas specific to Portkey

• Virtual keys → HiWay provider keys: Portkey's virtual key abstraction becomes your entries in Settings → Providers. Same encryption-at-rest posture, simpler mental model — no separate header to send on each request.
• Guardrails are out of scope: Portkey ships a moderation / content-filter layer that runs inline. HiWay doesn't — use OpenAI's Moderation API, Anthropic's prompt-level safety features, or a dedicated service like Lakera. Bring your own moderation if that was load-bearing.
• Prompt library doesn't exist on HiWay: Portkey lets you version prompts server-side and reference them by id. HiWay doesn't have an equivalent — manage prompts in your own code, or use a library like LangChain / Vercel AI SDK prompt templating.
• Fallback config → automatic same-tier fallback: Portkey's fallback config block is replaced by HiWay's default behavior (retry the next cheapest same-tier model on upstream errors, max 2 retries). If you had ordered fallback chains across specific models, pin with model: "<provider>/<id>" and handle retries client-side.
• Budget and alerting move too: Portkey's budget limits become Dashboard → Budget Control (monthly cap with BLOCK / DOWNGRADE / LIGHT_ONLY verdicts), and alerting becomes Slack / email webhooks on Guardian fires and budget transitions.