Deploy AI agents to the edge or the cloud — one API, any runtime.
Part of the [&] Protocol stack · Ecosystem overview · Three-protocol stack · Stack status
Welcome to the documentation hub for WebHost.Systems — a multi-runtime AI agent deployment and hosting platform built on Supabase, Cloudflare Workers, and AWS Bedrock AgentCore.
WebHost.Systems serves two roles:
Hosting platform for developers — create agents, deploy code bundles,
manage secrets, view logs and metrics, and handle billing through a single dashboard and API.
Hosting layer for the [&] Protocol ecosystem — validate ampersand.json
manifests, orchestrate MCP sidecars (Graphonomous, TickTickClock, etc.), route agent-to-agent skills via A2A, and enforce Delegatic governance.
WebHost.Systems is a Node.js/TypeScript monorepo in the [&] Protocol ecosystem. Built entirely on Supabase + Vite/React + Cloudflare/AWS runtimes.
A practical quickstart that gets a local dev environment running with Supabase, Vite, and Edge Functions
Architecture documentation covering the control plane / data plane split, RPI abstraction, and request flows
Skills guides for every operational surface — agent management, deployment, invocation, billing, security, and [&] Protocol integration
Spec references pointing to the authoritative engineering spec in docs/spec/
| Document | Description |
|---|---|
| Quickstart | Local dev setup, first agent, verification |
| Architecture | Control plane / data plane design, system diagram |
| Skills Reference | Operational skills index for agents and engineers |
Homepages
Root Docs
WebHost.Systems Docs
Skills
If you're new to the project, follow this path:
quickstart — local dev setup, first agent creation, verification
architecture — control plane / data plane design, request flows, data model
skills/SKILLS — skills registry and tool inventory
skills/01_AGENT_MANAGEMENT — agent CRUD, status lifecycle, provider config
skills/02_DEPLOYMENT_PIPELINE — bundle upload, validation, immutable deployments
skills/03_INVOCATION_GATEWAY — request/response shapes, auth, streaming
skills/04_RUNTIME_PROVIDERS — RPI abstraction, Cloudflare and AgentCore adapters
skills/05_TELEMETRY_AND_METRICS — metrics schema, aggregation
skills/06_BILLING_AND_LIMITS — subscription tiers, enforcement, checkout
skills/07_SECURITY_AND_SECRETS — RLS isolation, Vault secrets, audit
skills/08_AMPERSAND_INTEGRATION — [&] manifests, MCP sidecars, A2A routing
skills/09_ANTI_PATTERNS — common mistakes and how to avoid them
For operations and DevOps, start at skills/02_DEPLOYMENT_PIPELINE and skills/04_RUNTIME_PROVIDERS.
For [&] Protocol integration, start at skills/08_AMPERSAND_INTEGRATION and the 70_AMPERSAND_PROTOCOL_INTEGRATION.md spec.
The Supabase-backed layer that manages all platform state — agents, deployments, secrets, billing, and telemetry. Built on PostgreSQL with RLS for per-tenant isolation, Edge Functions for server-only logic, and Vault for encrypted secret storage.
Where agent code actually runs. WebHost supports two runtime providers behind a single abstraction:
| Runtime | Strengths | Best for |
|---|---|---|
| Cloudflare Workers + DO | Global edge, strong economics, instant cold starts | Most agents (default) |
| AWS Bedrock AgentCore | Long-running sessions, enterprise isolation, built-in tools | Premium/enterprise workloads |
The abstraction layer that makes runtime portability possible. The RPI defines a consistent contract for deployment, invocation, session management, and telemetry across all providers. Deploy once, run anywhere.
Deployments are append-only records. You never mutate a deployment — you create a new one. Rollback works by changing the active pointer, not by reverting code. This gives you a full audit trail and instant rollback.
Every invocation emits authenticated metrics — requests, tokens, compute milliseconds, errors. Telemetry feeds both the dashboard and the billing engine, so usage tracking and limit enforcement are built in, not bolted on.
A concise framing for the platform:
Supabase manages the control plane. Cloudflare and AgentCore run the data plane. The RPI makes them interchangeable.
The goal is not to build another LLM provider, but to provide a runtime-portable hosting layer where AI agents deploy, scale, and meter under a single API — regardless of which cloud actually executes the code.
The authoritative engineering spec lives in docs/spec/:
ADR-0005 — Deployment Immutability + Active Deployment Pointer Model
ADR-0006 — Canonical Invocation Protocol (invoke/v1) and Streaming
ADR-0007 — Tier Entitlements, Runtime Gating, and Limit Enforcement
ADR-0008 — Delegated Invocation Auth Mode (Server-to-Server HMAC)
WebHost.Systems/
apps/
web/ -- Dashboard frontend (Vite + React + Supabase Auth)
control-plane/ -- Supabase backend (Edge Functions + RPC)
supabase/ -- Supabase project (migrations, Edge Functions, seed)
packages/ -- Shared packages
docs/ -- This documentation
spec/ -- Authoritative engineering spec
skills/ -- Operational skills referenceMaster spec: Engineering Specification
ADRs: docs/spec/adr/
[&] Protocol integration: Ampersand Integration
[&] Protocol ecosystem: AmpersandBoxDesign/