// AI AGENT INFRASTRUCTURE

NexusLayer

Eleven services that give your code agent memory, routing, secrets, observability, knowledge, pipelines, review, remote control, and a team — wired together.

// what it is

Infrastructure for
agents that remember.

Your code agent — Claude Code, Cursor, Copilot, or your own — is a smart loop. It reads, thinks, writes. But every session starts from zero. NexusLayer changes that.

Eleven focused services, each solving one problem agents have in production: where do I store memory between sessions? which LLM should I call and how much did it cost? where are my API keys? how do I review code automatically? how do I control my agent from the other side of the world?

They share a single identity layer — one JWT, one login — and speak plain HTTP. Your agent calls them the same way it calls any API. No SDK required.

11
Focused services — each solves exactly one problem
1
Shared JWT identity across all products
∞
Agent memory — persistent across restarts and sessions
3
LLM providers — Claude, Gemini, Ollama — via one endpoint

// the ecosystem

Click any product
to see what it receives.

ModelRouter
LLM Gateway

Routes all your LLM calls to Claude, Gemini, or Ollama through one local proxy. Swap providers without changing a single line of your agent code.

POST router.nexuslayer.eu/v1/messages

your agent sends →

# Anthropic Messages API — unchanged
{
  "model":      "claude-sonnet-4",
  "max_tokens": 1024,
  "system":     "You are...",
  "messages":   [{
    "role":    "user",
    "content": "..."
  }]
}
# Router picks the active provider
# Your code never changes

// data flow guide

What your agent sends
to each product.

Real HTTP calls, real payloads. No SDK, no magic — just plain JSON over HTTP, authenticated with your shared JWT. Each card shows the minimal body your agent needs to send.

// product wiring

How the products
talk to each other.

These are the built-in flows: when you use one product, it can automatically feed another. Your agent doesn't orchestrate this — the platform does.

WatchGrid → ModelRouter Every LLM call routed through ModelRouter is automatically logged in WatchGrid with token counts and cost.
AgentVault → ModelRouter ModelRouter fetches provider API keys from AgentVault at startup — keys never appear in environment files or config.
AgentBrain → BrainVault Long-form notes saved via AgentBrain episodes are searchable in BrainVault — brain stores events, vault stores documents.
FlowMesh → AgentShop Pipeline nodes can dispatch tasks to AgentShop when automated steps need a human-supervised coding agent to execute work.
PeriodAI → WikiLLM PeriodAI report templates can query WikiLLM as a source — reports are grounded in your actual knowledge base content.
WatchGrid → FlowMesh WatchGrid alert rules can trigger FlowMesh pipelines — e.g. "if cost exceeds $5 in one run, start the on-call pipeline."
AgentShop → AgentBrain Agents running in AgentShop write their session summaries back to AgentBrain — the whole fleet shares one memory.
All products → AgentVault Every product fetches its service credentials and inter-product API keys from AgentVault at boot — one secret store for the whole platform.
Review → ModelRouter Review routes all its LLM analysis calls through ModelRouter — you switch the underlying model once and every review improves automatically.
Review → BrainVault Completed reviews are saved as BrainVault notes — your agent can query past findings to catch repeated mistakes across sessions.
TaskRelay → WatchGrid Every task dispatched through TaskRelay opens a WatchGrid run — you see token cost, tool calls, and duration for every remote session in the dashboard.
TaskRelay → AgentBrain When a remote session ends, TaskRelay writes the session summary back to AgentBrain — so the agent's memory is continuous whether you're home or abroad.