Built by makers, for makers.
Tokenwise is the LLM observability tool we wished existed when our own AI side-projects started costing real money. Plug it in once, get a weekly email that pays for itself.
Why we built this
Helicone went into maintenance mode. Langfuse is great but takes a weekend to set up. LangSmith is locked to LangChain. Meanwhile every maker shipping with the Vercel AI SDK or the OpenAI client was burning $200–$2,000/month on tokens with no idea where the money went.
So we built a proxy that adds < 50 ms of edge latency, captures cost / latency / errors, runs cheap caching and model-routing rules, and emails you a weekly digest of where to save money without hurting quality. One line of setup. No framework lock-in.
What we believe
- Optimisation should be one-click, not a research project. Recommendations land in your inbox with the savings already estimated and a button to apply them.
- Latency is part of the product. A proxy that adds 200 ms to every call is a tax, not a tool.
- Your data stays yours. Payload storage is opt-out per workspace; everything at rest is encrypted. See our security page for the details.
- $19 is a real price.Indie tier is meant to be cheaper than the bug you’d catch with it. We don’t do “contact sales”.
Who’s behind it
A small founding team in France, shipping in public. We answer every email at [email protected] — usually within a day. No support tier, no ticket queue.
Where we are
As of May 2026, Tokenwise has shipped the multi-provider proxy (OpenAI, Anthropic, Gemini, Grok, OpenRouter, Groq, DeepSeek, Mistral), workspaces with role-based access, alerts, evals, a semantic cache at the edge, weekly insights emails, public REST API, and an Optimize page with rules + A/B traffic splits.
We’re aiming for $5–10k MRR by November 2026 and 80k organic visits/month, mostly from programmatic SEO. If that doesn’t happen, we’ll change the plan. We won’t change the users.
Contact
Email: [email protected]
Status: tokenwisehq.com/status
Security disclosure: [email protected]