Skip to content

One ecosystem.
Four products.

The CLI learns from every task. The Router learns from every request. The LLM learns from every trajectory. The SDK exposes it all programmatically. A self-improving flywheel.

/ Products(01)
Open Source

Brainstorm CLI

AI coding assistant with intelligent multi-model routing. 42+ tools, 5 roles, 10 MCP servers, 4-mode TUI.

24 packages · Apache 2.0

SaaS

BrainstormRouter

AI API gateway with 13 intelligence systems. SPIFFE identity, streaming guardrails, semantic cache, evidence ledger.

362 models · 30+ providers

Research

BrainstormLLM

Sequential predictor trained on 2,203 trajectories. Predicts optimal 9-phase development pipelines.

0.796 F1 · <2ms ONNX

Library

@brainstorm/sdk

Programmatic TypeScript SDK. Run agents, classify tasks, analyze codebases from code.

npm · TypeScript

/ How They Connect(02)
CLI→RouterRoutes tasks to optimal models
Router→LLMPredictive routing from trajectory data
CLI→SDKSame engine, programmatic access
Router→Providers362 endpoints across 30+ providers
LLM→CLITrajectory data trains the predictor
/ The Flywheel(03)
1

Developer uses Brainstorm CLI to build a feature.

2

CLI routes the task through BrainstormRouter to the optimal model.

3

Router records the outcome — which model, what cost, did it succeed.

4

BrainstormLLM trains on accumulated trajectories, improving predictions.

5

Next time, routing is smarter. The system compounds.