Skip to content
← Back to blog
By JJ & Claude

Introducing Brainstorm: Why Every AI Task Deserves the Right Model

Announcing Brainstorm CLI — an open-source AI coding assistant that routes every task to the optimal model across 10+ models and 8 providers.

Most developers pick one AI model and use it for everything. A refactor gets the same model as a haiku-writing prompt. A security audit gets the same model as a commit message. This is like using a sledgehammer for every nail, screw, and picture hook in your house.

Brainstorm changes this. It is an open-source, CLI-first AI coding assistant that intelligently routes every task to the optimal model — across 10+ models and 8 providers — so you stop overpaying for simple tasks and stop under-powering complex ones.

The Problem with Single-Model Development

When you use one model for everything, you are making a silent trade-off on every single request. High-capability models like Claude Opus or GPT-5.4 are brilliant for architecture decisions, but they are slow and expensive for "rename this variable." Smaller models are fast and cheap, but they hallucinate on multi-file refactors. There is no universal best model — only the best model *for this task, right now*.

How Brainstorm Routes

Brainstorm's router uses Thompson sampling — a multi-armed bandit algorithm — to learn which model performs best for each task type. It starts with heuristic classification (is this a code generation task? a Q&A? a review?) and then continuously learns from quality signals. Over time, it converges on the optimal model for every category of work you throw at it.

The routing layer supports 6 strategies: quality-first, cost-first, combined, capability-based, learned (Thompson), and rule-based. You can switch between them with a single slash command.

What Ships in the Box

Brainstorm is not just a router. It is a full coding assistant:

  • 42+ built-in tools — filesystem operations, git integration, GitHub workflows, web search, task management, and agent orchestration
  • 5 roles — Architect, Engineer, Reviewer, Writer, and Explorer — each with curated model preferences
  • 10 MCP servers — extend Brainstorm with any Model Context Protocol server
  • 4-mode terminal dashboard — chat, dashboard, models, and config views in a rich TUI
  • Memory system — 4 types of memory with auto-extraction so context persists across sessions
  • Plan mode — think before you code, with step-by-step execution tracking

Open Source, Open Core

Brainstorm CLI is open source under Apache 2.0. The intelligent cloud routing layer — [BrainstormRouter](/router) — is a SaaS product that provides Thompson sampling across 357 models, streaming security inspection, and community-learned routing patterns. You can use the CLI with your own API keys and local models, or connect to the router for the full experience.

Get Started

Install and start building in under a minute:

``bash

npx brainstorm chat

``

Or explore the [CLI documentation](/cli) to see everything Brainstorm can do.

We built Brainstorm because we believe the model selection problem is the most under-addressed bottleneck in AI-assisted development. Every task deserves the right model. Now it gets one.