Site structure
A site is a self-contained directory holding everything one gateway runs: configuration, project sources, and the published store the gateway actually serves. This page walks the whole tree.
The site folder, end to end.
my-site/ ├── thermite-server # optional — binary deployed into the site ├── config/ │ ├── server.toml # gateway config — identity, devices, features │ ├── license.json # license (or trial.json while on trial) │ ├── gateway.id # generated on first boot — gateway identity │ ├── node.toml # node-local settings (redundant pairs) │ └── gateway.tls.*.pem # generated — gateway-to-gateway mesh TLS ├── projects/ # authoring workspace — what engineers edit │ └── press-line/ │ ├── project.toml # manifest: id, slug, branding, auth │ ├── users.toml # project-local users │ ├── tags/ # [[tags]] TOML files, subdirs allowed │ ├── screens/ # one folder per screen, index.html each │ ├── transactions/ # SQL logging groups │ └── queries/ # named SQL queries ├── published/ # runtime store — what the gateway serves │ └── press-line/ │ ├── production/ │ │ ├── current # pointer: names the live snapshot │ │ └── snap-<id>/ # immutable snapshot of the project │ └── drafts/wip/… └── logs/ ├── thermite-server.log.<date> └── audit.jsonl
How the tools find the site.
The server and the CLI resolve the site root differently — worth knowing before you script either.
Uses --site <path> when given; otherwise the directory containing the executable. A binary dropped into the site folder runs with no arguments.
Uses --site <path> when given; otherwise walks up from the current directory until it finds config/server.toml. Running commands from anywhere inside the site just works.
projects/ is what you edit. published/ is what runs.
The gateway never serves your working files. It serves immutable snapshots from published/, written by thermite project draft and publish — each one validated at publish time, recording the source git commit and branch on a best-effort basis when the workspace is a Git repo, and swapped in atomically via the current pointer. Source edits are served after publishing. To roll back project content, restore the earlier source revision and publish it again. The lifecycle has its own page: Publishing & the CLI.
What belongs in git.
Version source in projects/: manifests, tags, screens, transactions, and queries. Keep reviewed templates such as config/server.toml.example in Git. Keep live configuration, credentials, identities, licences, and private keys in a secured recovery set. Thermite installs source-oriented ignore defaults when a site has no .gitignore, including these rules:
# Selected rules from the installed site .gitignore
/published/
/logs/
/config/server.toml*
!/config/server.toml.example
/config/backup.toml*
/config/node.toml*
/config/gateway.id*
/config/gateway.tls.*
/config/license.json*
/config/trial.json*
/projects/**/users.toml*
!/projects/**/users.toml.example
.env*
*.key
*-key.pem
Preserve config/gateway.id and its matching config/license.json together in the secured recovery backup. Existing ignore files are preserved; defaults do not detect secrets or untrack committed files. Add rules for custom credentials_file and private-key names, and review templates before committing.
One site, many projects.
- One site = one gateway. A site can hold any number of projects, each an independent application with its own tags, screens, users, and roles.
- Project slugs and ids are unique across the site — the slug is the URL mount (/p/<slug>/) and the id keys sessions and tag scopes. thermite project init generates both.
- Tag paths are scoped per project — two projects can both define plant/Temperature without colliding.
- Devices and databases are site-level (declared in server.toml); a project's manifest scopes which of them it may use — see Projects.