Known limits

Last updated: October 7, 2026

FluxOS is in private beta while we harden the core. This page lists what doesn't work yet, so you can decide what to trust it with. We update it as fixes land.

Durability on FluxOS Cloud

• A commit is acknowledged once it reaches the workspace's local disk, before it is copied to durable storage. If the machine running a workspace is lost (out of memory, host failure), up to about five minutes of acknowledged writes in the default database can be lost.
• Databases other than the default one, and branches, are copied to durable storage only when the workspace suspends, or not at all for branches. A lost machine can take every write to them since the workspace last woke.
• Scale-to-zero suspends idle workspaces. In some failure cases a suspend can go ahead without its final checkpoint, with the same losses.
• Scheduled automatic backups don't run yet. A backup taken from the console covers only the default database, and point-in-time restore can return a state that never existed.

Until these are fixed, keep your own copy of anything you can't recreate.

Connecting to FluxOS Cloud

The Postgres wire protocol (psql, drivers, ORMs and BI tools) doesn't connect to Cloud workspaces yet. Use the browser workbench or the HTTP API.

Self-hosted servers listen for Postgres clients on port 5432; password login needs TLS configured. Some drivers misread results: row counts from INSERT, UPDATE and DELETE come back in the wrong format, and some statements describe no columns before sending rows.

SQL compatibility

• INSERT … RETURNING inserts the row but returns nothing.
• CREATE TABLE accepts PRIMARY KEY, UNIQUE, NOT NULL, CHECK, DEFAULT and REFERENCES but doesn't enforce them.

Most ORMs and migration tools rely on both, so they don't work with FluxOS yet.

Features with known gaps

• Time travel: after compaction, queries as_of a time before an update or delete can return wrong or empty results.
• Change data capture: a destination that falls more than about a thousand events behind silently misses changes.
• Encryption at rest: Cloud workspaces rely on storage-level encryption only, and key rotation doesn't work yet on self-hosted servers.
• Audit log: the hash chain can't be verified yet.
• On Cloud: masking policies and GDPR erasure are refused.

Experimental subsystems

These are in the product and you can try them, but they have known correctness gaps. Don't rely on them for data you need.

• Branches (creating a branch from a branch, and merging)
• Lakehouse tables (Iceberg and Delta)
• Kafka and Kinesis connectors
• Machine learning (train model, predict) and vector search
• AI insights
• Virtual warehouses
• The marketplace
• Federation (mounting another FluxOS server's tables)
• Clustering and replication, including failover

Questions

Found something that isn't listed here? Email hello@fluxos.run. Security reports go to security@fluxos.run.