Guided Cloud Setup

The guided installer can provision a hosted AutoMem backend for you and capture its endpoint and API key automatically — no dashboards, no copy-pasting tokens between terminals. This page covers that Hosted Cloud path in depth: how each provider flow works, what the installer captures, and how to script it.

Reach it by running the installer and choosing Hosted Cloud at the first prompt:

curl -fsSL get.automem.ai | sh
# or, if Node 20.19+ is already installed:
npx @verygoodplugins/mcp-automem install

?  Where should AutoMem run?
❯  Hosted Cloud      InstaPods or Railway — guided deploy
   Local Docker      Clone AutoMem and start Docker Compose on this machine
   Existing Endpoint Use an AutoMem URL you already have

Choose Hosted Cloud for always-on memory that follows you across devices and machines. For an on-machine stack instead, see Quick Start → Local Docker and Docker & Local Dev.

---

Pick a provider

After choosing Hosted Cloud, the installer asks how to stand it up:

?  How should we stand up your hosted AutoMem?
❯  InstaPods                      open the setup page — it deploys AutoMem and emails your URL + key
   Railway (guided)              sign in with the railway CLI, deploy from the terminal, then auto-capture keys
   Other — I already have a URL + key   already deployed somewhere; just paste your endpoint + token

Provider	How it deploys	Billing	Best for
InstaPods	Opens the setup page; it deploys AutoMem and emails your endpoint + key	$15/mo flat (details)	Fastest hosted path, HTTPS + custom domains, zero infra
Railway (guided)	Signs in with the railway CLI and deploys the template from the terminal	Usage-based (details)	Terminal-native deploys, fine-grained control
Other	You paste an endpoint + token you already have	—	An AutoMem instance you already run

InstaPods

1. The installer opens the InstaPods AutoMem setup page. Choose the Grow plan ($15/mo flat) and complete checkout.
2. InstaPods deploys AutoMem and emails your API URL + key.
3. Paste them back at the prompt:

?  AutoMem API URL  ›  https://your-automem.instapods.app
?  AutoMem API key (leave blank if this endpoint does not require one)  ›  ••••••••

The installer verifies the endpoint’s /health before touching any agent config.

Railway (guided)

1. The installer signs you in through the railway CLI (it installs/uses the CLI and opens a browser to authenticate).
2. It deploys the AutoMem template straight from the terminal, then auto-captures the endpoint and token once the service is healthy.
3. If the CLI can’t finish in your environment, it falls back to a browser deploy and then asks you to paste the URL + key, exactly like the InstaPods flow.

Railway requires PORT=8001

The AutoMem service must run on port 8001. The Railway template sets PORT=8001 for you. AutoMem’s own default already falls back to 8001 when PORT is unset (Flask’s bare 5000 default is never reached), but Railway assigns its own PORT per service — so other services can only reach memory-service reliably if PORT is set to 8001 explicitly.

Reuse vs. fresh deploy

The installer has a reuse path — if a provider reports existing AutoMem deployments, it lists them and lets you reuse one (fetching that deployment’s credentials) instead of paying for a new deploy. Any billable deploy is gated behind an explicit confirmation that names the plan.

Reuse is not yet wired up for Railway

In the current release the Railway provider reports no existing deployments, so the reuse selector is skipped and every guided Railway run deploys fresh. Reliable reuse-detection needs to enumerate the account’s projects and match the template source, which is deferred. If you already have an AutoMem instance running, choose Other — I already have a URL + key and paste its endpoint + token instead of deploying again.

---

What the installer captures

Once the backend is healthy, the installer:

1. Captures the endpoint + token from the provider (or from what you pasted).
2. Writes them to a .env in the current directory (AUTOMEM_API_URL, plus AUTOMEM_API_KEY if the endpoint needs one).
3. Wires your agents — review the plan, approve, and it registers the MCP server for each selected agent, backing up every file it changes with a .bak.

Nothing is written until you approve the plan, so a cloud run is as safe to preview as a local one.

---

Customize and script it

Every prompt has a flag and an environment variable, so you can pre-answer the cloud questions or run the whole thing unattended.

# Deploy to a hosted provider, fully interactive
curl -fsSL get.automem.ai | AUTOMEM_INSTALL_TARGET=cloud sh

# Pick the provider up front
curl -fsSL get.automem.ai | \
  AUTOMEM_INSTALL_TARGET=cloud \
  AUTOMEM_CLOUD_PROVIDER=instapods sh

# I already deployed — capture an endpoint without provisioning
curl -fsSL get.automem.ai | \
  AUTOMEM_INSTALL_TARGET=existing \
  AUTOMEM_API_URL=https://memory.example \
  AUTOMEM_API_KEY=sk-... \
  AUTOMEM_YES=1 sh

# Preview the whole plan, write nothing
curl -fsSL get.automem.ai | AUTOMEM_DRY_RUN=1 sh

The same flags work on the npm package, e.g. npx @verygoodplugins/mcp-automem install --target cloud --cloud-provider railway.

Flag	Environment variable	Purpose
--target	AUTOMEM_INSTALL_TARGET	cloud, local, or existing
--cloud-provider	AUTOMEM_CLOUD_PROVIDER	instapods, railway, or other
--endpoint	AUTOMEM_API_URL	AutoMem HTTP API endpoint (for existing/other)
--api-key	AUTOMEM_API_KEY	Bearer token for authenticated endpoints
--clients	AUTOMEM_CLIENTS	Agents to wire after deploy: codex,claude-code,cursor,openclaw,hermes
--no-agent-install	AUTOMEM_NO_AGENT_INSTALL=1	Provision the endpoint only; skip agents
--yes / -y	AUTOMEM_YES=1	Apply the reviewed plan without prompting
--dry-run	AUTOMEM_DRY_RUN=1	Print the plan, write nothing

Headless and CI

The installer never infers --yes from the environment. CI, CODEX, and CLAUDE_CODE only suppress the animated splash — they do not approve the plan. Without a TTY and without --yes (or AUTOMEM_YES=1), it prints the plan and stops — an unattended pipe can never make unreviewed changes, and a cloud deploy is never triggered without confirmation. For CI, pre-deploy with --target existing and pass an endpoint + key rather than provisioning inside the pipeline.

---

Verify

Confirm the hosted backend is healthy:

curl https://your-automem-url/health \
  -H "Authorization: Bearer YOUR_AUTOMEM_API_TOKEN"

{
  "status": "healthy",
  "falkordb": "connected",
  "qdrant": "connected",
  "memory_count": 0,
  "enrichment": { "status": "running", "queue_depth": 0 }
}

"qdrant": "disconnected" (with "status": "degraded") is expected if you haven’t configured Qdrant — AutoMem degrades to graph-only mode. See Quick Start → Verify it worked for the full field reference and a first end-to-end memory test.

---

Troubleshooting

Symptom	Likely cause	Fix
Railway CLI sign-in stalls	No browser / restricted shell	The installer falls back to a browser deploy; finish there and paste the URL + key
ECONNREFUSED after a Railway deploy	Railway assigned memory-service a different port than other services expect	Set PORT=8001 explicitly on memory-service (Railway guide)
401 Unauthorized on /health	Wrong/missing token	Re-check the key the provider issued; provide it without a Bearer prefix
Installer can’t find your new deploy	Credentials not ready yet	Re-run the installer, choose Other — I already have a URL + key, and paste the endpoint + token
Charged for a second deploy	Guided Railway runs always deploy fresh — reuse-detection is not wired up yet	Use Other and paste the existing deployment’s URL + key instead of re-running the guided deploy

---

Next steps

• Connect more agents — Platform Installers and the Platform Guides.
• Deep deployment docs — InstaPods and Railway.
• Production hardening — Backup & Recovery and Health Monitoring.