Docs

FAQ

What is ObjectiveAI?

ObjectiveAI is a REST API (api.objective-ai.io) and platform for scoring, ranking, and simulating decisions using large language models.

Instead of generating free-form text, ObjectiveAI focuses on producing structured outputs:

  • scores
  • rankings

These outputs can be trained from data, reused, and composed into larger decision systems.

What does ObjectiveAI actually do?

ObjectiveAI lets you:

  • score inputs (for example: "how good is this?")
  • rank options ("which of these is best?")
  • simulate preferences ("how would this person or system decide?")

It does this using ensembles of LLMs with explicit, learnable weights.

What are the main features?

ObjectiveAI provides three core capabilities:

  • Chat Completions - standard LLM chat, with reusable model configurations
  • Vector Completions - turn LLM judgments into numeric score vectors
  • Functions - composable, trainable scoring pipelines (what most users use)

What is a Chat Completion?

A Chat Completion is a normal LLM chat request: messages in, text out.

The difference is that ObjectiveAI lets you call configured model definitions, which may include:

  • built-in prefix or suffix prompts (for personality or instructions)
  • decoding and provider settings

From the user's perspective, it behaves like a standard chat completion.

What is a Vector Completion?

A Vector Completion produces numbers, not text.

It works by:

  1. Running multiple Chat Completions (one per model in an ensemble)
  2. Forcing each completion to produce a structured vote for a potential response. Each response is user-provided.
  3. Combining those votes using explicit weights
  4. Returning a vector of scores that sums to 1. The winning response is the one with the highest score.

Vector Completions are used for:

  • ranking options
  • selecting winners
  • producing machine-usable scoring outputs

What is a Function?

A Function is a reusable scoring pipeline.

Conceptually:

input -> score(s)

A Function:

  • accepts structured input
  • runs a list of tasks
  • each task is either a Vector Completion or another Function
  • produces either:
    • a single score, or
    • a vector of scores

Functions can be simple or complex, public or private, and are immutable and versioned.

Why would I use a Function instead of Chat or Vector Completions directly?

Because Functions are simpler and safer.

With Functions:

  • you don't manage prompts per request
  • you don't wire together ensembles yourself
  • you don't handle voting logic
  • you don't interpret raw vectors manually

You just send input and get back a score or ranking.

Can Functions be trained?

Yes.

Functions can be trained by providing example inputs and desired outputs.

ObjectiveAI:

  • keeps the underlying models fixed
  • learns weights over ensembles
  • caches AI outputs so optimization is CPU-bound
  • produces a Profile, which captures the learned behavior

This avoids fine-tuning models while still adapting behavior to your use case.

What is a Profile?

A Profile is a trained configuration for a Function.

It defines:

  • which ensembles are used
  • what weights are applied
  • how the Function behaves in practice

Profiles are:

  • immutable
  • versioned
  • reusable
  • shareable

A Function always has a default Profile, and you can publish new ones as you train.

Are you asking LLMs to produce scores?

No.

LLMs are not asked to produce numbers at any stage whatsoever.

Instead:

  • each model produces a vote
  • votes may be probabilistic by leveraging logprobs
  • votes are combined externally
  • behavior is learned from data, not self-reported confidence

This keeps the system simple, robust, and model-agnostic.

What is an Ensemble?

An Ensemble is a collection of configured LLMs used together.

Important properties:

  • Ensembles are immutable
  • Ensembles do not contain weights
  • Any change creates a new Ensemble ID
  • The same Ensemble can support many different behaviors via different Profiles

What are weights?

Weights control how much influence each model has during voting.

They are:

  • explicit
  • external to Ensembles
  • static per execution
  • learned during training

Weights are never hidden or implicit.

What LLMs are supported?

ObjectiveAI supports any LLM available through its upstream providers.

Ensembles can mix:

  • different model families
  • different providers
  • different configurations

The system is intentionally provider-agnostic.

How much does ObjectiveAI cost?

ObjectiveAI uses a credits-based pricing model.

Costs depend on:

  • the models used
  • the number of tokens processed
  • the size of ensembles
  • whether training is involved

Pricing reflects upstream model costs plus a service fee.

For custom pricing, please contact us.

Who is ObjectiveAI for?

ObjectiveAI is for:

  • developers building ranking or scoring systems
  • teams simulating preferences or decisions
  • products that need consistent, reproducible AI judgments

You do not need to be an ML expert to use it.

It is decision infrastructure.

Where can I learn more?

Documentation and examples are available on our website, and the open-source components live on GitHub.

If you're interested in ranking, scoring, or simulating decisions at scale, ObjectiveAI is built for that.

Objective Artificial Intelligence, Inc.