Server overview
agenticml server provides a central synchronization target and shared web dashboard for AgenticML experiments. It runs self-hosted on your own infrastructure or as our managed multi-workspace service on AWS. To use the hosted service instead of running your own server, start with the Hosted service section.
Architecture
Section titled “Architecture”- Workspaces: All runs, metrics, API keys, and team memberships belong to an isolated workspace on a single server hostname. Dashboard views run at
/w/<slug>/. - Database storage: Server state (accounts, workspaces, keys, sessions, audit events, run metadata, committed file sizes, and metric summaries) lives in a SQL database: SQLite for self-hosted single-process servers, or PostgreSQL for multi-process deployments.
- Run files and objects: Experiment files and shadow Git lineage repositories live on disk (or Amazon EFS). Content-addressed blobs (code, media, copied artifacts) live on disk or in Amazon S3.
- Single-process vs multi-process: SQLite deployments acquire an exclusive lock on
<data>/server/server.lockand run as a single process. PostgreSQL deployments coordinate concurrency via database locks and support multiple server processes behind a load balancer. - POSIX requirement: The server runs on Linux and POSIX hosts (Python 3.10+). (The client CLI and
agenticml uisupport Windows). - TLS termination: The server speaks plain HTTP on its configured port. In production, terminate TLS using a reverse proxy such as AWS ALB, Tailscale, or Caddy.
Topics
Section titled “Topics”- Installation and deployment: Install via pip, optional extras (
[postgres],[s3],[firebase],[hosted]), run directly or under systemd, and configure a TLS proxy. - Authentication and keys: Workspaces, roles (
owner,admin,member,viewer), workspace-bound API keys, and dashboard sessions. - Syncing runs: Connect training machines with
agenticml login,sync --all, continuoussync --follow, and storage limit handling. - Backups and restore: Online consistent backups via
agenticml server backup, SQLite snapshots, PostgreSQL dumps, and restore. - Upgrades and troubleshooting: Upgrading to 3.0, automatic 2.x data migration into
default, and troubleshooting common operational issues.