1. Pilae
2. Apps
3. LibreChat

Running LibreChat in production: what it takes

1. Pin and deploy

   A LibreChat version we have run, with MongoDB, Meilisearch and the RAG API beside it, reachable only on your private network behind the Pilae gate.

2. Configure providers centrally

   Every endpoint and key lives in librechat.yaml and the secret store. People pick a model from the list you approved; nobody pastes a personal API key.

3. Wire sign-on and groups

   OIDC against your identity provider, local registration off, and access to expensive or external models limited to the groups that need them.

4. Back up conversations, agents and files

   MongoDB, the vector store and uploaded files go offsite nightly. Once a month we restore them into a scratch environment and open a restored conversation.

5. Upgrade in your window

   LibreChat releases often and its configuration schema moves with it. The Pilae Agent tests the next version against a copy of your configuration, writes the plan and applies it once you approve.

What LibreChat is, and who runs it

LibreChat system requirements

Before anything is deployed, this is what has to exist. We size it with you in the first session, and we say so when your own hardware is already enough.

CPU and memory

2 vCPU · 4 GBFor the interface, the API and the search index. Inference is elsewhere: on your GPUs or at a provider.

Database

MongoDBLibreChat stores users, conversations and agents in MongoDB. It is the one component of this stack that is not Postgres, and we back it up on its own schedule.

Search

MeilisearchFull-text search across a person's conversation history. Optional in principle; people expect it within a week.

Retrieval

RAG API · PostgreSQL with pgvectorFor files uploaded to a conversation or an agent. Embeddings can run on a local model so documents are not sent out to be indexed.

Model backend

OpenAI-compatible endpoint, or provider keysvLLM or Ollama on your GPUs, or keys for the hosted providers you allow. Each one is configured once, centrally.

Identity provider

OIDC, SAML or LDAPSo access follows your directory instead of a list of local passwords.

Migrating from ChatGPT Team to LibreChat

ChatGPT Team gives each person one vendor and one bill per seat. LibreChat gives them the same kind of interface over several providers at once, and the move is mostly a decision about which providers your organisation allows. Conversation history can come across: each person exports their ChatGPT data and LibreChat imports the file into their own account. Custom GPTs do not convert. Their instructions and files are copied into LibreChat agents by hand, which is also a good moment to retire the ones nobody opens. The API keys move from individual accounts to one place you control, so a leaver takes nothing with them.

1. Decide which providers are allowed

   Local models, hosted APIs, or both, and for which groups. This is a data decision before it is a technical one, and it is written into the configuration rather than into a policy document.

2. Put it behind sign-on

   OpenID Connect against Keycloak or Entra ID, with email registration switched off. Accounts arrive from your directory and leave with it.

3. Import history and rebuild the GPTs that matter

   Each person imports their ChatGPT export into their own account. The custom GPTs still in use are rebuilt as LibreChat agents with the same instructions and files.

4. Run both for a month, then cancel the seats

   The old subscription stays live while usage moves. We report which models people choose, so the seat count you cancel is based on what happened.

Several providers, one configuration file

version: 1.2.1
cache: true
registration:
  socialLogins: ['openid']
interface:
  modelSelect: true
  agents: true
endpoints:
  custom:
    - name: 'Local (vLLM, zur1)'
      apiKey: '${VLLM_API_KEY}'
      baseURL: 'http://vllm.mesh.internal:8000/v1'
      models:
        default: ['chat-model']
        fetch: true
      titleConvo: true
      titleModel: 'current_model'
  anthropic:
    titleConvo: true
  agents:
    disableBuilder: false

What Pilae is responsible for

Where it runs

zur1, fra1, fal1, gra1, ams1, hel1, lon1, ash1, hil1, sin1, tok1, syd1, on-premZurich, Frankfurt, Falkenstein, Gravelines, Amsterdam, Helsinki, London, Ashburn, Hillsboro, Singapore, Tokyo, Sydney, Your own hardware

Who holds the credentials

You do. Ours are separate, named, logged and revocable with one command. We ask before anything changes outside an agreed window.

If you leave

The machine, the data, the compose files and the runbook are already yours. Nothing stops when our access does.

What running LibreChat costs

On Starter, LibreChat runs in Pilae Cloud for CHF 449EUR 479USD 559 a month, hosting and onboarding included. On Business, it is one of up to three apps for CHF 1,590EUR 1,690USD 1,990 a month, on your own servers or in Pilae Cloud. Enterprise is quoted. The plans set what each includes; these are the inputs that move the price around them.

Instance size

The CPU, memory and, where a model runs, the GPUs the app needs for your users and your data.

High availability

One machine with tested restores, or a replicated setup that keeps serving when a node fails.

Storage and backups

How much data it holds, how long backups are kept, and point-in-time recovery for its database.

Plan and support

Starter, Business or Enterprise: support hours, response times in the contract and how often we review the service with you.

Region

Your own hardware, where the infrastructure is already yours, or a Pilae Cloud region, where machines are priced per month. On Starter, Pilae chooses the site.

Sign-on and integrations

Single sign-on, directory sync, mail relays and the other systems the app has to reach.

LibreChat: common questions

Also in ai and llm