Skip to content
Tokenomix Request a demo

Platform

One hop between your apps and every model they call.

Tokenomix is a proxy. Applications point at it instead of at the provider, and it does five things to every request before anything leaves your network. The provider can be a vendor API or a model on your own GPUs; the gateway treats both the same.

  1. 01

    Identity

    The validated API key resolves to a person and a department.

    Never a client-supplied header. A caller can claim to be anyone; a key can only be the thing it is.

  2. 02

    Quotas

    Tokens, requests and spend checked against every limit that applies.

    Runs before screening, so a request that is over budget costs no GPU time. Most restrictive limit wins.

  3. 03

    Screening

    Seven pattern detectors plus a fine-tuned classifier.

    Jailbreaks, PII, secrets, prompt injection, toxicity, blocked topics and your own regex.

  4. 04

    Policy

    Block, warn, redact, log or allow. Strictest match wins.

    Department rules layer on top of central rules and can only ever tighten them.

  5. 05

    Audit

    Verdict, tokens, cost and the deciding rule, on every call.

    The row is written whether the request was allowed or stopped, which is what makes chargeback possible.

01 — The three jobs

Cost control and data control, in one place.

Most organizations buy these separately and end up with two sets of numbers that disagree. Running both off the same proxy means the spend report and the security report are built from the same rows.

Screening and policy

Catch what shouldn’t be in the prompt

Seven pattern detectors and a fine-tuned classifier, wired to a policy engine where the strictest matching rule wins.

Read on →

Chargeback and quotas

Give every dollar an owner

Spend attributed to a person, a team and an application. Then hard limits on tokens, requests and cost.

Read on →

Routing

Decide which model gets to see it

Constraints narrow the candidates, preferences pick one. A prompt carrying PII can be pinned to a local model.

Read on →

02 — The console

Where the day-to-day happens.

Admins see the whole estate. A department admin sees their own department and can only ever tighten what central IT set. An end user sees their own keys and their own spend, and nothing else.

Dashboard showing request volume over time, verdict breakdown, token usage by model and top consumers.
Volume, verdicts, token usage by model and the top consumers, over any window from an hour to a month.
Request insights showing traffic segments, which models serve which workloads and per-model cost in milliseconds per output token.
Insights reports per-model cost as milliseconds per output token. Raw latency mostly measures how much text was asked for.

03 — Fitting in

What it takes to put this in front of your traffic.

OpenAI-compatible
One base URL change. The chat completions and embeddings endpoints behave the way your SDK already expects, streaming included.
Hosted vendor APIs
OpenAI, Anthropic, Azure OpenAI, Google and Cohere. Keys live in the gateway rather than in forty application configs, encrypted at rest with a key you hold, and each provider’s own auth scheme is handled for you.
Models you run yourself
vLLM, Ollama or anything else speaking the OpenAI-compatible API, on-prem or in your own VPC. Same screening, same quotas, same chargeback rows. Self-hosted traffic is usually the half nobody is measuring, because it never shows up on an invoice.
Streaming
Streamed responses pass through and arrive the way they always did. The answer is screened for PII and secrets once the stream finishes, so a leak in a streamed reply still gets a verdict and an audit row.
Identity
Roles for admin, operator, viewer, department admin and end user. Personal API keys, minted and revoked per user, with the raw secret shown exactly once.
Audit
Every request row carries the verdict, the detector that fired, the deciding rule, tokens in and out, cost, and whether screening was bypassed.
Export
Chargeback downloads as CSV grouped by user, model, provider, department or application. Finance can take it straight into a spreadsheet.
Request a demo Deployment detail