Skip to content
Agents tracked: 444 Releases (7d): 644 Agent status: 3 with issuesSince last limit reset: Claude 31d · Codex 3d Updated Oct 23, 2026
AgentGid
✦ Ask

Model APIs and gateways · by Tokenwise Labs

Tokenwise

A drop-in edge proxy for OpenAI, Anthropic, Gemini, Grok, OpenRouter and other LLM APIs that logs cost, tokens, latency and errors per call and suggests fixes such as model swaps, caching and prompt trims, with rules for fallbacks, A/B splits and budget caps.

You find out which prompts and models drive your LLM bill by changing one base URL, while your own provider keys pass through and are not stored.

API (paid)Free tierCommercial OK

Open tokenwisehq.com ↗ agent.md

For your project

Official API
Official API · Paid Public REST API: 1,000 calls/hour on Indie, 10,000/hour on Pro; proxy plan caps 200,000 (Indie) or 2,000,000 (Pro) requests/month, hard stop at 120%
“Tokenwise is an HTTP proxy. Point your existing SDK at our base URL with your tw_live_* key — no client library to install, no SDK to maintain.” tokenwisehq.com/docs
Key
Needed
“Read-only access to your requests, metrics, evals via tw_api_* bearer tokens.” tokenwisehq.com/docs
Formats
JSON
Data download
Not stated
Price
Free tier · from $19/mo No free plan: 7-day trial without a card (200,000 requests/month, 30-day retention)
“Indie $19 / mo 200,000 / mo 60 days 10 Pro $79 / mo 2,000,000 / mo 180 days 50” tokenwisehq.com/terms
License
Tokenwise Terms of Service Use of the proxy, dashboard and public API
Commercial use
Commercial use OK
“Developers and small teams running LLM features in production: apps and SaaS that call the OpenAI, Anthropic, Google, or other model APIs” tokenwisehq.com/
Credit
Not stated
Embedding
Embedding unclear
Using the website
Paid

Read on Tokenwise's own pages on Oct 24, 2026 · license or terms · API docs · pricing. Check the license page before you ship.

For your agent

A request your code or agent can make (add your key):

https://tokenwisehq.com/api/status
  • Keep the OpenAI/Anthropic SDK and set baseURL to https://proxy.tokenwisehq.com/<provider path> (e.g. /openai/v1, /anthropic, /openrouter/api/v1) plus the Tokenwise key header (docs show x-tw-key, the home page X-Tokenwise-Key); read data via /api/v1/public/* with a tw_api_* bearer token.
  • Requests stop at 120% of the monthly plan limit; size the plan to traffic
  • Do not resell raw proxy access as a competing observability product
  • Prompt payload storage is on by default; turn it off per workspace or tag for sensitive data

Paste into your agent: Read https://agentgid.com/resources/tokenwise/agent.md before using Tokenwise.

Quick answers

Does Tokenwise have an API?

Yes. Tokenwise has an official API documented at https://tokenwisehq.com/docs.

Does the Tokenwise API need a key?

Yes, you need an API key. Limits: Public REST API: 1,000 calls/hour on Indie, 10,000/hour on Pro; proxy plan caps 200,000 (Indie) or 2,000,000 (Pro) requests/month, hard stop at 120%

Is Tokenwise free?

Yes, there is a free tier: No free plan: 7-day trial without a card (200,000 requests/month, 30-day retention) Paid plans start at $19 a month.

Can I use Tokenwise in a commercial project?

Yes. Its terms say: "Developers and small teams running LLM features in production: apps and SaaS that call the OpenAI, Anthropic, Google, or other model APIs"

What you can build

  • Per-feature LLM cost dashboard for a small SaaS by tagging each route's calls
  • A fallback chain that moves a chatbot to another provider on 5xx or 429 errors
  • A weekly report that pulls request metrics from the public API and flags cost spikes

Alternatives in Model APIs and gateways

All 7 →

Zro You can point an existing coding agent at cheaper open models by changing a base URL, without request bodies being stored or used for training. API (paid)Commercial OK Cohesor You can cut and cap agent model spend by swapping one base URL and key instead of rewriting code for each provider. API (paid)Commercial OK LLMTest A provider outage or a reply that is not valid JSON no longer breaks your feature, because the proxy retries on another model inside the same request, and you see what each AI feature costs. API (paid)Commercial OK Nous Research (Nous Portal inference API) You can call Hermes models through a standard OpenAI client, or download the same weights from Hugging Face and run them yourself. APIDownloadFree tierCommercial: conditions Opper AI Switching or pooling model providers becomes a model-string change, and teams that must keep data in the EU get one sub-processor and one DPA instead of a contract per provider. APIFree tierCommercial OK RunInfra You can point Claude Code, Codex, Aider or an existing OpenAI/Anthropic SDK at open models by changing a base URL and key, without renting or tuning GPUs yourself. API (paid)Free tierCommercial OK

Something wrong or out of date? Tell us.