Skip to content

Capacity and scaling

Flux is built for a team on one server. One Compose project runs one PostgreSQL, one API, one worker and one keyring on one Docker host, and that is the only deployment that is tested.

  • Memory. Raise FLUX_API_MEMORY (default 1536m), FLUX_WORKER_MEMORY (512m) and FLUX_DB_MEMORY (768m) in the settings file and apply them with ./flux up. A container that reaches its limit is killed and restarted; for the API that also locks every encrypted space. docker stats shows how close each one runs.
  • Database. Each API and worker process keeps a pool of 10 PostgreSQL connections, and PostgreSQL keeps its default of 100. Give the host fast disk before more CPU: every write ends with an ordered event insert. PostgreSQL itself runs with its image defaults; tune it through a Compose override file only if you measure a need (./flux reads docker/compose.override.yaml).
  • Disk. The database keeps every saved version. Watch the pgdata volume and keep room for a backup, which needs about twice its size while it is written.
  • Agent runtime. Everyone with Connect AI has a place; how many agents run at once follows from memory, about 2 on 4 GB, and the others wait (what each agent needs).
  • Live media. Calls run in LiveKit with a configured limit of 16 people per room; CPU and bandwidth grow with each call. That limit is configured, not a measured capacity.

The performance measurements of October 2026 record what was measured on a seeded data set, with the method.

Running several API or worker containers, PostgreSQL on another host, or Flux across several nodes is not supported in this version:

  • the keyring must be exactly one per deployment and reach api and worker over a private network on the same host;
  • api and worker share the files volume as a local directory;
  • nothing in the shipped files deploys, tests or documents more than one of each.

Several independent instances on one host are fine: each checkout gets its own Compose project, volumes and port. Watch Docker’s address pools when you run many (Docker address pools).