Changelog
Telemetry comes to shed: an optional stack with Grafana dashboards for every project and service, an OpenTelemetry gateway for your apps, and notifications to Hark, ntfy, or email when something fails. The dashboard has a new look, and shed no longer stores client IP addresses.
Features
Databases
- Added a metrics exporter and per-statement query statistics to Postgres, MySQL, MongoDB, and Redis while telemetry is on. A database deployed before this release is redeployed once, one at a time, when telemetry first turns on, and is unavailable for the few seconds the restart takes. (#12)
Observability
- Added an optional telemetry stack under Server → Telemetry: VictoriaMetrics, VictoriaLogs, VictoriaTraces, and Grafana run as containers on the server, listening on loopback only, and Grafana is served at
/grafana/behind the dashboard's sign-in. (#12) - Added default Grafana dashboards for shed, every project, and every service, with deploys, restarts, stops, crashes, backups, and restores marked as annotations. A dashboard you edit is never overwritten, and Restore default dashboards brings back missing ones. (#12)
- Added request metrics, access records, and a trace for every request the proxy serves, with no change to your apps. (#12)
- Added an OpenTelemetry gateway: while telemetry is on, each app's next deployment gets
OTEL_EXPORTER_OTLP_ENDPOINTand the otherOTEL_*variables injected, and its spans, metrics, and logs go tolocalhost:4318inside its own container. Per-service caps bound how much a service can send. (#12) - Added shipping of every service container's output to the log store, with searchable history on the Logs tab. (#12)
- Added a record of container events (
docker exec, OOM kills, restarts) and of each service's outbound connections, with a Security dashboard listing destinations a service reaches for the first time. (#12) - Added an audit log under Server → Audit recording every change, sign-in, denial, and agent tool call. It is always on, and agents can read it through the
audit_eventstool. (#12) - Added traces of shed itself while telemetry is on: deployments, builds, backups, restores, API and MCP requests, and outbound calls, with each deployment linking to its trace. (#12)
- Changed shed to never store client IP addresses: they stay out of the audit log, access records, and spans,
caddy.logdrops the client's address, headers, and URI, and panic logs are scrubbed. Upgrading removes the addresses already in the audit log. (#35)
Dashboard
- Redesigned the dashboard on Cloudflare's Kumo: a new look and logo, tabs that keep your place, and a ⌘K palette that searches projects, services, actions, and docs. Updates, backups, notifications, telemetry, and integrations moved under Server, and the site at shed.land matches, with link previews. (#37)
- Changed the projects list to sort by last updated. (94e4a53)
Agents
- Added telemetry tools to the MCP server: query metrics and logs, search and read traces, list a service's slow requests and a database's top queries, and list Grafana dashboards. (#12)
- Changed MCP audit events to record only IDs, times, and limits, never a query, span name, or tags, which could hold the secret being looked for. (41eb1ee)
Installation and Updates
- Added update channels under Settings → Updates: Stable offers stable releases and Nightly also offers nightly builds of
main. A shed installed from a nightly starts on the nightly channel, and the installer acceptsSHED_CHANNEL=nightly. (9d98131)
Other
- Added notifications under Server → Notifications: rules send a message to every enabled channel when a deploy, backup, or restore fails, a service runs out of memory or crashes in a loop, or a metric alert fires in Grafana. Messages are stored in shed.db and retried until they send. (#9)
- Added ntfy and SMTP email channels, alongside Hark push. (7f93ee0)
Bug Fixes
- Fixed every 1Password call failing with "out of bounds memory access" until shed restarted, after one request to 1Password was canceled midway. (9a478d8)
1Password now keeps every service's variables in one vault you choose, with one service account for all of shed.
Features
Variables
- Added a 1Password vault for variables. Create a vault in 1Password, then a service account with read and write access to it, and paste its token in Settings → 1Password. shed adds an item for each service, titled
<project> / <service>and tagged by project, and moves your variables into them. (#11) - Changed the Variables tab to read and write those items. Edits made in the 1Password app apply on the next deploy. (#11)
- Added automatic item upkeep: new services get an item, renaming a project renames its items, and deleting a service archives its item. (#11)
- Changed disconnecting 1Password to copy the variables back to the server. (#11)
Breaking Changes
- Removed the per-service 1Password Environment links from v0.2.4 and v0.2.5. Connect once in Settings → 1Password to move your variables into a vault. (#11)
- Added a 1Password vault for variables. Create a vault in 1Password, then a service account with read and write access to it, and paste its token in Settings → 1Password. shed adds an item for each service, titled
Each service's 1Password link now has its own service account token.
Features
Variables
- Changed each service's 1Password link to use its own service account token, so a token can read only that service's Environment. 1Password can't change a service account's access after it's created, so create one per Environment and paste its token with the Environment ID from the 1Password button on the Variables tab. (67d2631)
- Improved linking so shed reads the Environment before saving the link, and a wrong ID or token fails right away. (67d2631)
- Removed the 1Password card from Settings. (67d2631)
Bug Fixes
- Fixed deploys failing with "invalid client id" a while after 1Password was connected. (#10)
Breaking Changes
- Services linked in v0.2.4 keep their Environment but lose the shared token, so their deploys fail. Add a token for each on its Variables tab.
Services can now load variables from a 1Password Environment.
Features
Variables
- Added loading variables from a 1Password Environment. Connect a 1Password service account in Settings → 1Password, then link an Environment with the 1Password button on a service's Variables tab. shed reads the Environment each time the service deploys and shows its variables in the Variables tab. Variables you set in shed take priority. 1Password doesn't announce changes, so redeploy after you change a value. While connected, 1Password uses about 110 MB of shed's memory. (#10)
Features
Dashboard
- Improved the metric charts so they end at the limit when there is one (vCPUs, total memory and disk, a service's limits) and otherwise fit the data closely. Byte axes no longer overshoot, for example 46.6 GB for a 23.4 GB server. (01f6074)
Agents
- Changed Settings → Agents to show one connect command at a time, with tabs for any agent, Claude Code, Codex, and the URL. (d9170d0)
- Added
npx add-mcp(any agent) and Codex connect instructions to Settings → Agents, the MCP docs, and theshedskill, alongside Claude Code. (3bb4fe4)
- Added
Secrets in shed.db are now encrypted at rest, agents can inspect shed through a read-only MCP server, and shed has a public site at shed.land.
Features
Networking
- Added
proxy.cloudflare, which trusts Cloudflare's edge for the client IP and setsX-Forwarded-ForandX-Forwarded-Hostcorrectly for proxied domains. (#4)
Dashboard
- Added a new logo and favicon. (4d1f305)
- Improved the layout on phones with a full-bleed frame. (6903049)
- Removed the right-hand outline from docs pages. (a51fb74)
Agents
- Added a read-only MCP server at
/mcpthat lets agents like Claude Code inspect projects, deployments, build and runtime logs, and metrics. Agents sign in through a built-in OAuth 2.1 server, and Settings lists and revokes them. The repository ships ashedskill for agents. (#6)
Installation and Updates
- Added a public site at shed.land with the docs, and a shorter installer:
curl -fsSL https://shed.land/install.sh | sudo bash. (#5)
Other
- Added encryption at rest for secrets in shed.db: variables, GitHub App credentials, the backup age identity, S3 secret keys, and deployment runtimes. On first start shed generates
/etc/shed/shed.key(or reads it from systemd credentials) and converts the existing database. (#7)
Breaking Changes
- Back up
/etc/shed/shed.keyoff the server. Restoring shed.db needs it, and shed refuses to start with the wrong one.
- Added
Features
Variables
- Changed the Variables table to hide values until revealed, and copying a value with references now copies it resolved. (#3)
Dashboard
- Removed the image's
@sha256:digest from the service header. (#3)
Bug Fixes
- Fixed logs masking variable values shorter than 8 characters, which mangled timestamps and JSON in services with values like
1ortrue. (#3)
The first release of shed.
Features
Deployments
- Added projects and services: apps from a GitHub repo (Dockerfile or Railpack) or a Docker image, and one-click Postgres, MySQL, MongoDB, and Redis. (#1)
- Added push-to-deploy through a GitHub App, optionally waiting for CI, with zero-downtime switchovers, rollbacks, and build and runtime logs. (#1)
Variables
- Added variables with
${{ service.KEY }}references. (#1)
Networking
- Added domains with automatic HTTPS, volumes, resource limits, and public TCP ports. (#1)
Backups
- Added scheduled, compressed, optionally encrypted backups to local disk and S3, with restore and download. (#1)
Observability
- Added host and service metrics and shed's own log. (#1)
Installation and Updates
- Added a one-command installer, signed releases, and in-app updates. (#1)
Other
- Added in-app docs. (#1)