Genosyn Enterprise adds SSO and the Audit log.
A self-hosted install runs Community edition with unlimited AI Employees, unlimited Routines, and no key to enter. An Enterprise license turns on two features. It changes nothing else about the software you already have.
Audit history is recorded on Community too. The pages that read it show an available-in-Enterprise card instead of the trail, so nothing is missing on the day a key is activated. On Genosyn Cloud the same two features arrive with the Scale plan.
Features a signed key turns on. Nothing else about the software changes.
AI Employees
Routines
Single sign-on
Audit log
One string beginning genlic1., signed Ed25519 and checked against public keys compiled into the software.
There is no activation server and no phone-home. An air-gapped install validates the same key the open internet does.
A master admin pastes it at Admin / License. No restart and no rebuild.
A paid key expires soft: the features stay on past the date and the status card shows a renewal warning. An evaluation key expires hard.
The seat count is informational. Genosyn records it, shows it beside the number in use, and never blocks a hire over it.
Removing the key returns the install to Community edition. SSO and the Audit log switch off and nothing is deleted.
Genosyn is one container listening on port 8471.
Everything that has to survive a restart is either a database row or a file under /app/data. Model calls go to the endpoints you registered. Connections reach the accounts you authorized.
A diagram: a boundary labelled your network, holding the Genosyn App on port 8471, its database on SQLite or Postgres, the AI Models you registered, the Connections your Grants scope, and the data volume. The license issuer sits outside that boundary with no connection to it.
Souls, Skills, Routines, Run logs
Database rows on sqlite or postgres. Both drivers carry every entity and every migration.
Model keys, Connection credentials, SSO client secret
Encrypted at rest with AES-256-GCM and never returned to the browser after they are saved.
Git checkouts, browser state, attachments
Files under /app/data, which the installer maps to the named volume genosyn-data.
Browser recordings
Silent MP4 files under data/.private/browser-recordings, capped at 2 GiB each. They never enter an employee working tree.
Instance secrets
data/.instance-secrets.json at mode 0600, with a matching key ID in the database so a missing file stops startup instead of being replaced quietly.
The system interposing on an action the employee already attempted. A human ticks it and the server replays that exact call from the snapshot on the row. Approving fires a privileged side effect, so it is admin-gated and the payload is redacted at every boundary. Six kinds ship today.
The employee choosing to stop and ask. It writes the question and the options itself, answering one performs no side effect, and an ordinary Member can answer it. Anything privileged the employee does afterwards still meets its own Approval.
A revocable stop on all AI work at one scope, placed by a human or tripped by the consecutive-failure breaker. Runs already moving finalize interrupted rather than failed, because nothing failed. Queued retries keep their due time and fire after the lift. A slot that arrives during one is declined and the schedule advances, so lifting an old Standdown produces no catch-up storm.
Under those three, sixteen Grant tables decide what one AI Employee can reach: Connections, Repositories, mail accounts, calendars, Vault items, Member browsers, and ten more. A Check is the machine-verifiable assertion a Run must pass before it finalizes green, and the graded employee cannot author one. There is deliberately no tool that lets the roster place a Standdown, and far more importantly, none that lets it lift one.
Three supported topologies start at one Docker host.
Pick the one that matches what your team already operates. The database driver and the model authentication decide the rest of the shape.
One Docker host
One replica, SQLite in the data volume, the container on 8471 behind whatever reverse proxy you already run for TLS. Bubblewrap isolates coding and repository work. Where user namespaces are unavailable, boot falls back to disabled and everything except the coding tools still runs.
Kubernetes
Kubernetes 1.27+, a 20Gi ReadWriteOnce volume at /app/data, external Postgres, and your own Ingress. The pod turns Ready only after every migration has run: /api/health answers once boot completes, so a pending probe in the first minute is normal.
Shared, Postgres-backed
Several replicas coordinate through Postgres leases, database-backed auth flow state, and cross-replica realtime fan-out. The volume becomes ReadWriteMany. ChatGPT subscription access stays on one trusted single-tenant App process; API-key and custom models scale with the rest.
CLI installs run genosyn upgrade daily at 03:17 local. The previous container is kept until the new one is ready, and a failed start brings the old version back on the current data volume.
Archives land in data/Backup/, and can be mirrored to an SMB or SFTP destination whose credentials are encrypted with the same helper as model keys.
Off until you set a day count at Admin / Backups. Two things survive it whatever their age: the newest completed archive, and any archive you uploaded yourself. Copies already delivered off-box are never touched.
5 consecutive bad Runs put a Standdown on that Routine, recorded with source breaker rather than a person. 0 switches it off.
Priority support ships with the Enterprise license.
Community support is GitHub Issues, read by the people who wrote the code, and it stays free. A license adds a direct line and four pieces of work you would otherwise do alone.
The topology, the database driver, volume sizing, ingress, and the upgrade path, read against the environment you already run rather than a reference one.
A data-flow map for your controls: what is a database row, what is AES-256-GCM at rest, what never leaves /app/data, what an employee token can write, and what the license verifies without a network.
SSO client registration, the callback URL derived from your public URL, and whether Create accounts on first sign-in stays on. Password login keeps working either way, so resetting SSO cannot lock an operator out.
A restore rehearsal from data/Backup/, the per-driver migration stream, the containment threshold for your risk appetite, and the first upgrade watched with you.
None of it is a gate on the software. Every AI Employee, Routine, Integration and section of the product is identical in Community, and the source is on GitHub under Apache 2.0 for anyone who would rather read it than ask.
Enterprise questions go to enterprise@genosyn.com.
Four lines are enough to get a useful answer back. You get a topology, the questions a security review usually asks, and a price if you want one.
One Docker host, a Kubernetes cluster, or a shared Postgres estate.
Google, an OIDC provider, or email and password for now.
Where the volume lives, who backs it up, and what your retention window is.
Which roles the AI Employees would hold, and what a Run of theirs would touch.