Skip to content
Layer 02 — Artificial Intelligence

Brello AI

Generative intelligence system

What it does

Capabilities

Generative design across modalities

Brello AI produces structured candidate designs across text, code, schemas, and visual specifications. The model is fine-tuned for design completeness and constraint satisfaction rather than for surface-level fluency.

Strategic decomposition

Given a high-level objective, Brello produces a tree of subordinate questions, dependencies, and decisions — explicitly marking which choices materially change the outcome and which are downstream consequences.

Plan execution at agent scale

Brello is designed to issue tasks to agentic systems while retaining oversight. Plans are produced with explicit checkpoints, observability primitives, and interrupt protocols.

Architecture

How it is built

  1. 01A reasoning core trained for plan composition, with structured output formats native to the model.
  2. 02A constraint-satisfaction layer that runs both inside the model and at runtime, enforcing user-specified invariants.
  3. 03A plan-tracing layer that emits the model’s decision graph in machine-readable form for inspection and replay.
  4. 04An oversight layer with explicit interrupt and rollback primitives.
Use cases

Where it fits

Designing a system architecture

A senior engineer asks Brello for a system architecture under a stated set of constraints. Brello returns three candidate architectures, an explicit comparison along the constraint axes, and the next questions whose answers would change the recommendation.

Planning a complex launch

A product team gives Brello the launch goal, the available resources, and the hard constraints. Brello produces a phased plan with named decision points, observability requirements, and explicit fallbacks per phase.

Decomposing a research program

A research lead gives Brello a multi-quarter research goal. Brello returns a decomposition into experiments and milestones, with explicit hypothesis-falsification gates, and the model is willing to say "this is unlikely to work" when the evidence supports it.

Roadmap

Status & milestones

  1. ShippedQ3 2025

    Internal usage

    Used across Apik teams for plan composition and design review.

  2. In progressQ2 2026

    API preview for partners

    Limited partner program with rate-limited access.

  3. PlannedQ4 2026

    Public API and Studio surface

    Self-serve API, web Studio interface, and structured-output SDKs.

FAQ

Common questions

  • Is Brello a wrapper around an existing model?

    No. Brello is a model trained from scratch on plan-composition and design-completion data, with significant architectural choices that diverge from a pure language model.

  • Can Brello drive my agents?

    Yes. Brello is designed as the reasoning + planning layer for our agentic infrastructure. It can also drive partner agent stacks via standard protocols.

  • How does Brello handle uncertainty?

    Plans include explicit uncertainty annotations and decision-relevance markers. The model is trained to surface uncertainty rather than hide it.

  • What about safety?

    Brello’s outputs are governed by the Apik Acceptable Use Policy. Frontier-capability releases are gated by our Responsible Development Policy.

  • How is Brello priced?

    API access is metered. Studio access has subscription tiers. Detailed pricing will publish with the public API.

  • Where does Brello fit in the stack?

    Brello is Layer 02 — Artificial Intelligence. Layer 03 (Agentic Systems) operates on top of it.

Related across the site