Repository navigation
Releases
v0.3.1-nightly.1
This release contains only routine updates to shed's dependencies, with no changes you will notice.
v0.3.0
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)
v0.2.7-nightly.9
Notifications can now go to ntfy and email, and the projects list shows what changed most recently.
Features
Dashboard
- Changed the projects list to sort the most recently updated projects first. (94e4a53)
Other
- Added ntfy and SMTP email as notification channel providers alongside Hark, set up under Server → Notifications. (7f93ee0)
v0.2.7-nightly.8
The dashboard is redesigned on Cloudflare's Kumo design, with browser-style tabs, a new shed logo, and a Server page that now includes what Settings used to. Deployment and backup lists page through their whole history.
Features
Deployments
- Changed the deployment history to page 10 at a time through the whole history instead of stopping at the 50 newest. (#37)
- Added an Active deployment card to the Deployments tab: what is serving now, when it went live, where it came from, and how long it took. (#37)
Backups
- Changed the backup lists to page 10 at a time, past the 100 newest shown before. (#37)
Dashboard
- Redesigned the dashboard on Cloudflare's Kumo design, with flatter surfaces, a new shed logo, and Open Runde as the typeface. (#37)
- Added browser-style tabs to the top bar: keep several places open at once, pin, close, and drag them into order, open links in a new shed tab with ⌘-click or a middle click, and find them back between visits. (#37)
- Changed the Server page to show sections beside a summary of the machine: Metrics, Logs, and the Audit log under Monitor, and Updates, Backups, Notifications, Telemetry, and Integrations under Configure. (#37)
- Changed the project page to a diagram of the project's domains, services, and the services they use, with the latest deployments below. (#37)
- Rebuilt the metric charts as interactive charts with a legend and hover tooltips. (#37)
- Improved ⌘K search to find server sections alongside projects, services, actions, and docs pages. (#37)
Installation and Updates
- Added a button to the Installation docs that copies a prompt asking a coding agent to install shed over SSH. (#37)
Other
- Added a changelog page and link previews to shed.land, and redesigned its home page. (#37)
Breaking Changes
- Removed the Settings page. Updates, agents, 1Password, and backup storage are now sections of Server: check for updates under Server → Updates. (#37)
- Changed the deployment and backup list API responses to pages:
GET /api/services/{id}/deploymentsreturns{ deployments, next }instead of an array, and the backup lists gained anextcursor. Pass the previous page'snextas?before=to read on. (#37)
v0.2.7-nightly.7
This release contains only internal changes to CI and linting, with nothing new for operators.
v0.2.7-nightly.6
shed no longer writes client IP addresses to disk: not in the audit log, not in access records, not in caddy.log, and not in the spans it receives.
- Removed the client address from the audit log: events no longer record where a request came from, the page under Server → Audit no longer shows it, and upgrading deletes the addresses already in
shed.db. (#35) - Removed the client address, query string, user agent, and referer from access records, and the "Top client IPs", "Sign-ins by IP", and "Edge top client IPs" panels from the default Grafana dashboards. (#35)
- Removed the client IP from the
slow_requestsMCP tool. (#35) - Removed client addresses, request headers, and URIs from
caddy.log, and scrubbed the client address from the server's and the proxy's panic logs. (#35) - Dropped the client address and user agent attributes from OTLP spans shed receives; outbound spans keep them, since their peer is the server being called. (#35)
Breaking Changes
- Client addresses are gone from audit events and access records, and no setting brings them back. Update log queries and Grafana panels that use
client_ip, and API clients that read an audit event'sactorIp. Backups ofshed.dband telemetry data from before the upgrade keep their addresses until retention removes them.
v0.2.7-nightly.5
shed can now push notifications to your phone when a deploy, backup, or restore fails, when a service runs out of memory or crashes in a loop, or, with telemetry on, when a metric crosses a threshold.
Features
Observability
- Added notification channels, with Hark push as the first provider; the webhook URL is treated as a secret, encrypted in
shed.dband never returned by the API, and Send test verifies a channel on demand. (#9) - Added event rules that notify on failed deploys, backups, and restores, on out-of-memory kills, and on crash loops (three crashes within 10 minutes by default), all working with telemetry off. (#9)
- Added metric rules for host disk and memory use, HTTP 5xx ratio and p95 latency, database connections, and dropped telemetry, evaluated by Grafana with thresholds and durations editable in shed. (#9)
- Added a
shedcontact point and notification policy in Grafana, so alert rules you create there notify the same channels. (#9)
Dashboard
- Added a Notifications tab under Server with channels, rules, and the newest 50 deliveries with their status, attempts, and last error; failed sends retry for up to six hours. (#9)
Breaking Changes
- Removed the provisioned Grafana alert rules for service restarts, OOM kills, and failed deploys and backups; their alerts no longer appear on the Telemetry page. Add a channel under Server → Notifications to get these events pushed to you.
v0.2.7-nightly.4
This release only adds development tooling: shedlab runs throwaway shed installs in local VMs to test shed itself. Nothing changes on a server running shed.
v0.2.7-nightly.3
- Improved the telemetry gateway to rebuild the masks that keep variable values out of stored telemetry in the background, so an ingest request no longer waits while shed re-reads every variable from 1Password. (259f512)
v0.2.7-nightly.2
- Fixed deploys and the Variables tab failing with "out of bounds memory access" until shed restarted, once any request to 1Password was cancelled midway. (9a478d8)