Layerr is a 12-layer orchestration intelligence system that sits between your applications and LLM providers. A dedicated router model classifies every prompt into 8 intent categories, then a workload analyzer, workspace selector, 12-dimensional scoring engine, and strategy engine determine the optimal model. Fallback chains, provider health checks, explainability, replay intelligence, adaptive learning, and benchmark harnesses are all built in. Currently v0.8.0-alpha - early but real.

Does Layerr generate AI responses?

No. Layerr never generates responses itself. It acts as an intelligent routing and orchestration layer, forwarding your prompts to existing providers like Anthropic, OpenAI, Mistral, Google, and others, then returning the response back to your tool.

Is it compatible with Claude Code and Cursor?

Yes. Layerr exposes a fully OpenAI-compatible API endpoint. Set OPENAI_BASE_URL to Layerr and Claude Code, Cursor, Roo Code, or any OpenAI SDK client routes through it instantly. Zero code changes. The author has been running this against Claude Code for weeks - local models for quick patches, escalated to sonnet for heavy reasoning.

How does intelligent routing work?

Layer 1: a router model classifies intent into CODING, REASONING, CREATIVE, VISION, DOCUMENT, FAST, SECURITY, or GENERAL. Layer 2: deterministic workload analysis scores complexity, latency sensitivity, cost sensitivity, context size, and reasoning depth with zero LLM calls. Layer 3: a workspace profile (General, Coding, Research, Creative, Local-First, Cost, or Performance) is selected. Layer 4: the strategy engine builds a 12-dimensional scoring profile with latency tolerance, cost sensitivity, reasoning depth, and fallback thresholds. Layer 5: every candidate model is scored across 12 dimensions plus runtime enrichment. Layer 6: the primary model is selected and a fallback chain of up to 4 candidates is prepared. If the primary fails, the next candidate is tried automatically. Every decision produces a structured explainability output.

Can I self-host Layerr?

Yes. The Layerr runtime deploys via Docker or Kubernetes on any cloud or bare metal. Self-hosted deployments use your own API keys, run all routing logic locally, and receive all 9 capabilities including workspace customization, replay intelligence, benchmark harness, and entitlements. In v0.8.0-alpha the focus is on getting the local + cloud hybrid right first.

What models are supported?

Any model available through OpenRouter (100+ models), plus direct integrations with Anthropic, OpenAI, Mistral, Google Gemini, Cohere, and Ollama for local inference. The router is currently qwen2.5:14b or similar. If a provider exposes an API, Layerr can route to it.

Is my code or data sent to Layerr?

Only routing metadata is processed - estimated prompt complexity, context size, and task type. Your actual code and prompts go directly to the chosen LLM provider. Self-hosted deployments send zero data to Layerr servers. The router model runs locally if you self-host.

What's the difference between Layerr and OpenRouter?

OpenRouter is a model marketplace - one API key, many models. Layerr is an orchestration intelligence system that decides WHICH model to use, WHEN to escalate, and HOW to recover from failures. It adds intent classification, workspace profiles, 12-dimensional scoring, fallback chains up to 4 deep, explainability, replay intelligence, adaptive learning, benchmark harnesses, and governance. OpenRouter gives you access. Layerr gives you decisions.

Security

Security architecture

Self-hosted infrastructure changes the security model entirely.

Self-hosted guarantees

Because Layerr runs on your own infrastructure, the security guarantees are different from a SaaS product. Your data never leaves your perimeter unless you choose to send it to an LLM provider. There is no multi-tenant database to breach. There is no Layerr admin panel with access to your prompts. The attack surface is your own infrastructure, which you already control.

API key handling

API keys are stored encrypted at rest in your local configuration. They are never logged in plaintext. They are never transmitted to Layerr servers because Layerr does not operate servers that receive your runtime data. Keys are used only to authenticate with the LLM providers you configure.

No proxy, no middleman

Layerr is not a proxy service. Your traffic goes directly from your infrastructure to the LLM providers you configure. The routing decision happens locally. The request is forwarded from your machine to Anthropic, OpenAI, or whichever provider you selected. There is no Layerr server sitting in the middle.

Threat models we address

The current architecture addresses: prompt interception by a third-party middleware, vendor lock-in through proprietary APIs, and data residency concerns by keeping everything on your infrastructure. It does not yet address: formal compliance certifications, advanced threat detection, or enterprise-grade access controls. These are on the roadmap as the product matures beyond alpha.

What is still in progress

Layerr is v0.8.0-alpha. Formal SOC 2, ISO 27001, and compliance work are not yet in scope. They will be added as the product matures and as enterprise customers require them. The security architecture described above is the foundation, not the finished product. If you have specific requirements, we want to hear about them.