Latitude self-hosting: requirements and options

the short answer

Yes. Latitude is MIT-licensed and self-hosts for free: Docker Compose or Docker Swarm on one host, a Helm chart on Kubernetes, or a Railway template, all deploying the same images as Latitude's cloud. You run five application services on Postgres with pgvector, ClickHouse, Redis, Temporal and S3-compatible object storage, and you own backups, upgrades and TLS. The managed alternative runs in AWS eu-central-1, Frankfurt.

Licence: MIT, since May 2026
Options: Docker Compose, Docker Swarm, Helm, Railway
Stack: 5 app services; Postgres + pgvector, ClickHouse, Redis, Temporal, object storage
Managed cloud: AWS eu-central-1, Frankfurt

The Deployment Options

Latitude (latitude.so) moved its repository from LGPL-3.0 to MIT in May 2026, and its deployment docs state the principle plainly: "Every self-host option deploys the same images with the same configurable values." You are not running a community edition; you are running the product.

option	shape	notes from the docs
Docker Compose	One machine	"A production-grade instance on one machine"
Docker Swarm	One host, scaled	The stateless services scale horizontally; bundled infrastructure stays single-instance unless you swap in managed services
Helm chart	Kubernetes	A deployment per service; web, API and ingest exposed through the ingress, workers and workflows internal only
Railway	Managed PaaS	A pre-configured deploy button

Deployment options from Latitude's documentation.

What You Run

Five application services - web, api (which also serves MCP), ingest, workers and workflows - plus the infrastructure under them:

PostgreSQL with the pgvector extension, for application data.
ClickHouse, for telemetry. This is where traces land and what grows with volume.
Redis, for caching and queues.
Temporal, the workflow engine.
Object storage: SeaweedFS, S3-compatible, bundled.
A one-shot migration job that sets up the databases.

On Kubernetes every piece is replaceable. The Helm chart bundles Postgres, ClickHouse, Redis, Temporal and SeaweedFS, and each can be turned off with <dependency>.enabled: false and pointed at a managed service in the external: block. If your company already runs managed Postgres and ClickHouse, that removes most of the operational weight.

What You Operate

Backups. On Compose, persisted state lives in the postgres_data, clickhouse_data, seaweedfs_data and redis_bullmq_data volumes; on Kubernetes, in the equivalent PVCs. Snapshot them or use each database's own tools.
Upgrades. Pin image tags in .env.production; migrations run automatically on restart. On Helm, helm upgrade runs migrations as pre-upgrade hooks.
TLS and exposure. Put a reverse proxy - Caddy, nginx or Traefik - in front of an internet-facing Compose install. The Helm chart expects three public domains, for web, API and ingest, with TLS terminated at the ingress controller.
Secrets. The docs warn that the generated secrets "are permanent for the installation". Keep that file, or move the keys into a pre-created Secret, before anything else.
Retention. Telemetry retention is set by LAT_TELEMETRY_RETENTION_DAYS, which defaults to 3,650 days. Ten years of traces in ClickHouse is rarely what anyone meant; set it on day one.

LAT_TELEMETRY_RETENTION_DAYS=90

In .env.production: keep 90 days of traces instead of the 3,650-day default.

When Self-Hosting Beats the Cloud

Latitude's managed cloud stores and processes customer data in AWS eu-central-1, Frankfurt, and names no other region. Self-host when one of these is true:

Your data has to stay somewhere else - in the US, in your own cloud account, or inside a network with no egress. Self-hosting is the only way to run Latitude outside Frankfurt.
Your volume is high. Pro includes 100K credits and charges $20 per extra 10K. At high volume, infrastructure you already pay for can undercut the meter; a credit is not defined on the pricing page, so estimate from a trial.
You already run the stack. A team with managed Postgres and ClickHouse and some Temporal experience is mostly adding stateless services.

Stay on the cloud when none of that applies. Starter is free with unlimited seats, Pro is $99 a month, and neither needs anyone on call for ClickHouse. If you need on-prem with Latitude's own support and an SLA, that is the Enterprise tier. Prices are on Latitude pricing.

What Self-Hosting Does Not Change

Self-hosting changes where Latitude runs, not how its workflow behaves. It still provides tracing, completed-session evaluations and monitors. Teams that also need automated analysis across sessions, findings with recommended fixes or runtime policies need an additional workflow around that observability data.

If you are choosing between self-hostable platforms, the footprints and workflows differ. Langfuse self-hosts under an MIT core with Docker Compose or Kubernetes. Future AGI uses an Apache-2.0 core and a larger multi-service stack. Failproof AI Cloud self-hosts on Enterprise and combines tracing and querying with evaluations, automated failure analysis, findings and alerts. Its MIT CLI runs tested behavioral policies locally. See self-hosted agent evaluation platforms for the full comparison.

FAQ

Is self-hosted Latitude free?

Yes. The latitude-llm repository is MIT-licensed, with no licence fee and no credit meter when you run it yourself. You pay for the infrastructure - Postgres, ClickHouse, Redis, Temporal, object storage and five application services - and for the time to operate it. Supported on-prem deployment is part of Latitude Enterprise.

Is self-hosted Latitude the same as Latitude cloud?

Latitude's deployment docs say every self-host option deploys the same images with the same configurable values as its cloud, and the platform is MIT-licensed in full. What differs is who operates it, and which region the data lives in: Latitude's cloud runs in AWS eu-central-1.

Can I use managed databases with self-hosted Latitude?

Yes. Latitude's Helm chart bundles Postgres, ClickHouse, Redis, Temporal and SeaweedFS, and each can be disabled with its enabled flag and replaced by an external managed service in the chart's external configuration block. On Docker Swarm, bundled infrastructure stays single-instance unless you swap in managed services.

Get Started

Failproof AI is free to start. It finds recurring failure modes across agent sessions using code-based and LLM-based evaluations, groups the evidence into findings, and recommends fixes. Bring the eval suite you already have, alert the right owner when behavior drifts, and turn a tested fix into a policy that prevents the failure from recurring. See pricing for the tiers.

start free get a demo →

Sources

Checked against each vendor's own site and docs on 2026-09-14. Products change; if a detail here is out of date, tell us at support@befailproof.ai.