# MargIQ

> MargIQ learns recurring request patterns inside production AI workflows. It uses lower-cost models from your available set where evidence supports them and keeps stronger models where quality or risk demands it.

MargIQ is not a generic LLM router and is not tied to a specific model provider. It is a decision layer for AI workflows. It starts with workflow evidence, shows potential savings and safety behavior, and lets teams control each workflow with Automatic, Report only, or Disabled modes.

## Best Starting Points

- [Homepage](https://getmargiq.com/): Product overview and positioning.
- [Product](https://getmargiq.com/product): Workflow routing, safety guards, and control plane.
- [Docs](https://getmargiq.com/docs): Developer documentation.
- [Quickstart](https://getmargiq.com/docs/quickstart): Add MargIQ without changing production model behavior.
- [npm package](https://www.npmjs.com/package/margiq): Official MargIQ SDK package.
- [LLM Setup](https://getmargiq.com/docs/llm-setup): Copy-paste prompt for AI coding assistants.
- [Examples](https://getmargiq.com/docs/examples): Routine workflow examples.
- [Proof](https://getmargiq.com/proof): Evidence model and validation approach.
- [Controlled benchmark](https://getmargiq.com/benchmarks/workflow-aware-llm-routing): Path-level cost reduction and quality-protection evidence across a controlled 15-workflow suite.
- [Benchmark data (JSON)](https://getmargiq.com/benchmarks/workflow-aware-llm-routing.json): Machine-readable methodology, measured routes, quality protection, and limitations.
- [Benchmark summary (Markdown)](https://getmargiq.com/benchmarks/workflow-aware-llm-routing.md): Agent-readable benchmark evidence and citation.
- [Pricing](https://getmargiq.com/pricing): Free and Pro plan differences.

## Product Summary

- Free plan: report-only mode, estimated potential savings range, basic workflow signals, no active backend routing policies.
- Pro plan: request-level routing paths inside each workflow, analyzer-backed recommendations, policy learning, safety guards, and per-workflow controls.
- Trust model: show where MargIQ saves money, where it refuses to optimize, and how each workflow can be controlled.
- Developer setup: install MargIQ in the backend/server-side AI module, keep existing provider credentials unchanged, add only MargIQ API credentials, pass `provider` and `availableModels`, and keep the normal provider response shape.

## Language To Use

Use: workflow, routing path, optimized model, safety guard, Automatic, Report only, Disabled.

Avoid: workflow hashes, variant hashes, rule IDs, embeddings, thresholds, raw policy JSON.

## Contact

- Email: support@getmargiq.com