Migrate off Helicone
Move from Helicone to Tokenwise in one line
Same one-line proxy idea you already use — but still actively shipping, and instead of just dashboards you get a weekly email with the exact changes that cut your bill.
Why people are moving
Helicone was acquired by Mintlify in March 2026 and is now in maintenance mode — the proxy still runs, but it isn’t getting new models, providers, or workflow features. If you want a tool that keeps pace with LLM pricing, you need one that’s still being built.
Helicone popularized the one-line proxy: change your baseURL, get observability. Tokenwise keeps that exact mental model, so migrating isn’t a project — it’s swapping a URL and a header. Your app code, your SDK, and your provider keys all stay the same.
The difference shows up after setup. Where Helicone stops at a dashboard you have to read, Tokenwise closes the loop: a weekly email naming the specific model swaps, cache rules, and fallbacks that lower your bill — each one checked for quality and verified against your own traffic before it counts.
Helicone vs Tokenwise
| Tokenwise | Helicone | |
|---|---|---|
| Setup | One line — swap your baseURL | One line — swap your baseURL |
| Actively shipping | Yes — new features weekly | Maintenance mode since the Mintlify acquisition |
| New models & providers | 8 providers, pricing kept current | Frozen — no new providers |
| Weekly action email | Yes — the changes to make this week | No |
| Cost recommendations | Applied at the edge, savings verified on your traffic | Dashboards only |
| Edge actions | Semantic cache, fallback chains, model overrides, A/B | Caching + rate limiting |
| Maker pricing | $19/mo Indie, $79 Pro | Free, then a jump to $79 Pro |
| Open-source / self-host | Cloud only | Open-source (Apache 2.0) |
The actual change
You’re already pointing your client at a proxy baseURL with an auth header. Migration just changes which host and header you use — nothing else in your code moves.
Before — Helicone
import OpenAI from "openai";
const openai = new OpenAI({
baseURL: "https://oai.helicone.ai/v1",
defaultHeaders: {
"Helicone-Auth": "Bearer sk-helicone-...",
},
});After — Tokenwise
import OpenAI from "openai";
const openai = new OpenAI({
baseURL: "https://proxy.tokenwisehq.com/openai/v1",
defaultHeaders: {
"X-Tokenwise-Key": "tw_...",
},
});- 1
Create a workspace and key
Sign up, create a workspace, and copy your labeled proxy key (tw_…). Takes about a minute.
- 2
Swap the baseURL and header
Point your baseURL at https://proxy.tokenwisehq.com/openai/v1 and send your key in the X-Tokenwise-Key header instead of Helicone-Auth. Your provider key stays exactly where it is.
- 3
Send one request
Your first call appears in the dashboard via live polling — cost, latency, tokens, model, status, all tracked immediately.
- 4
Read your first weekly email
Within the week you get the Weekly Insights email: the specific changes that cut your bill, each quality-checked before it's suggested.
What you keep
- Your provider keys. Tokenwise is BYOK — your OpenAI/Anthropic keys ride with the request and never persist on our infrastructure.
- Your code.It’s a baseURL change, not an SDK migration. No wrappers, no decorators, no LangChain.
- Your data, exportable. Everything we hold is available through the public REST API — no per-row export fees.
- An exit. Repoint the baseURL back in seconds if you ever want to leave. Low switching cost in means low switching cost out.
Frequently asked
Is Helicone shutting down?
No — the proxy still works. But after the Mintlify acquisition in March 2026 it moved into maintenance mode, so it isn't getting new models, providers, or workflow features. Tokenwise is the actively-shipping, maker-priced option with the same one-line setup.
How long does migration take?
A few minutes. You change the baseURL and swap the auth header from Helicone-Auth to X-Tokenwise-Key. Nothing else in your application changes.
Do I have to give you my OpenAI key?
It's BYOK: your provider key travels with each request to the upstream provider and is never written to our disk, logs, or cache. You can also disable payload storage per workspace if you only want metrics.
What about the latency of routing through a proxy?
The proxy runs at Cloudflare's edge with under 50ms of overhead, and logging is async fire-and-forget so it never blocks the provider response. Same-shape fallback chains mean a degraded upstream still completes your request.
Can I keep self-hosting like I did with Helicone?
Tokenwise is cloud-only today — that's the honest trade. In exchange you run nothing: no Postgres, no ClickHouse, no proxy to operate. If self-hosting is a hard requirement, Helicone's open-source build or Langfuse may fit better.
It’s a one-line change
Point your baseURL at Tokenwise, keep your keys and your code, and get a weekly email with the changes that cut your bill — proven on your own traffic before we recommend them.