ModelMesa

Pre-launch · Early access open

Govern every decision call. Starting with Jev.

Approved-model policy, per-team keys, spend caps, audit logs, and PII detection, all in front of the Jev key you already have. Flat subscription, no per-token markup.

$ curl -s $MODELMESA_URL/v1/recommend \
    -H "Authorization: Bearer mm_live_…" \
    -d '{"task": "Summarize this ticket"}'

{
  "model": "fast-mid",
  "confidence": 0.81,
  "probabilities": {
    "fast-mid": 0.86,
    "frontier-large": 0.11,
    "small-cheap": 0.03
  },
  "policy": "support-default@prod"
}

Illustrative example. The API may change before launch.

What you get

The control plane your API key is missing

Approved-model policy

List the models your company has approved. For each task, ModelMesa picks the best fit from that list, never outside it, and returns the per-model probabilities and the confidence behind the pick. Low confidence falls back to your default.

PII guardrails

Detect personal data in requests to Jev and redact it from stored logs, using a detector based on OpenAI's Privacy Filter that we run ourselves. Mask and block come in v1. Set policy for the whole org, a project, or a single call. Org-level floors stay in force.

A data-minimization aid, not a compliance guarantee.

Billing, usage, and logs

One portal for developers, ops, and finance, organized by company. See usage and estimated cost by project, key, and model, set spend alerts, and search request logs with redaction and retention controls.

Bring your own key

Keep your existing account, contract, pricing, and limits. Your key is envelope-encrypted, decrypted only by the gateway at request time, never shown back, and never resold. Works with your TypeSafe Jev API key.

  • Security-first gateway
  • No per-token markup
  • Project-scoped, rotatable keys
  • Audit trail for keys and policies
  • Jev-compatible request bodies

How it works

From raw key to governed API in four steps

  1. Connect your key

    Add your existing API key in the portal. It's encrypted on arrival and never displayed again.

  2. Point your app at ModelMesa

    Send your Jev requests to the ModelMesa endpoint with a project-scoped mm_live_ key. Your request bodies stay the same.

  3. Set policies

    Approved models, PII rules, rate limits, and spend caps for the whole org or per project. PII rules can also be set per call.

  4. Watch usage

    Track usage, estimated cost, and logs per project, and get an email alert when spend crosses a threshold you set.

Be first in line

ModelMesa is in development. Join the early-access list and we'll reach out when there's something to try.