InferenceIndex

Docs

What a Compiled Agent Model does

Use a Compiled Agent Model when the same agent job comes back with new inputs.

A Compiled Agent Model is the repeatable part of work an agent has already figured out. InferenceIndex learns from the first agent runs and compiles what repeats into a Compiled Agent Model: a reusable way to carry out the same kind of work on new inputs.

The first time, the full agent does the work. The next time the same kind of task shows up, with a new ticket, invoice, lead, or other input, the Compiled Agent Model carries it out. Checks stay in place. If the learned path no longer fits, the run goes back to the full agent.

When to use it

Use a Compiled Agent Model when all of these are true:

  • The same kind of task comes back. New inputs, same job.
  • A full agent can already do that job once.
  • You want the next runs to finish sooner and cost less than paying for the long run every time.

Jobs that fit, as published on the InferenceIndex home page:

  • Support ticket triage. The same 30 kinds of tickets, sorted and tagged the same way, all day.
  • Invoice reconciliation. The same match, with new vendors, amounts, and dates.
  • Lead follow-up. New lead, same research, same kind of first touch.
  • Meeting notes to CRM. The same summary, next steps, and field updates.
  • Inbox sorting and replies. A handful of request types with known answers.
  • Weekly reporting. The same metrics, sources, and write-up.
  • Onboarding new customers. The same accounts and welcome steps, a new name.
  • Content repurposing. One post becomes the same set of follow-on posts.

When not to use it

Skip it when the task is new each time, when you only need one answer, or when a full agent cannot do the job yet. New work still goes to the full agent. A Compiled Agent Model does not replace that path, and it does not guess.

What you get

The outcome, again, on the next input. The home page shows early repeat runs as 5–10× faster, and says repeat runs are often 10-100x cheaper on our tests. Savings depend on the workflow.

Where to go next

Request access on the waitlist. Once you have access, call a Compiled Agent Model from an OpenAI-compatible client: swap your base URL and model name.

  • Install and quickstart: https://inferenceindex.com/docs/quickstart
  • Pricing in plain terms: https://inferenceindex.com/docs/pricing
  • Home and waitlist: https://inferenceindex.com/#waitlist