CHANGELOG.md

Changelog

[0.4.0-beta.7] - 2026-08-23

Seventh beta of the Albacore release, focused on reliable first deployments through the CLI.

Added

• Bind mount ownership declarations and file ownership changes
• One-shot service jobs whose status and output follow the container process
• Partial environment variable updates with plan support
• OpenAPI metadata for accepted values and plan capable operations

Changed

• Deployment environments use .env as the principal file while retaining .env.flatrun compatibility
• Environment discovery reports every .env file and identifies the principal file
• Deployment images report the resolved image alongside its compose source expression
• Registered database connections provide defaults for database browsing
• PostgreSQL and postgres are accepted as database type names

Fixed

• Seeded bind mounts retain image ownership, and redeploys no longer apply one service owner to another service's mounts
• PostgreSQL users can create tables in databases provisioned for them
• Routed services join the proxy network before a domain is reported as configured
• Failed domain changes restore deployment metadata and compose configuration
• Deployment and service jobs time out instead of blocking later actions indefinitely
• Nested settings updates preserve omitted sibling values
• Multi-database deployments do not receive misleading legacy database variables
• Compose validation can read environment values supplied in the same create request
• Environment based image updates keep the compose expression and update its variable

[0.4.0-beta.6] - 2026-08-23

Sixth beta of the Albacore release, making connected servers manageable as one Fleet.

Added

• Guided Fleet setup, peer access policies, remote deployment inventories, and runtime provider selection
• Host and deployment capacity decisions with managed horizontal and vertical scaling
• Docker Swarm and k3s orchestration adapters with nginx and Traefik routing adapters
• Grouped incidents and configurable notification targets and delivery rules
• HTTP, TCP, and container command health checks for web services and databases
• Deployment management grants for administering allowed deployments through a peer
• Service-specific health checks for mixed web and database deployments

Fixed

• Existing Fleet peers gain default access policies during startup repair without reconnecting
• Existing peer credentials are restricted to their configured Fleet policy during startup
• Fleet peer credentials can read deployments allowed by their peer policy
• Fleet readers can open deployment details without gaining write access
• Object storage and notifications have independent permission boundaries
• Updates require dedicated access and remain admin-only by default
• Settings, notifications, and API keys require explicit access for non-admin roles
• Repeated metric alerts share one incident until every affected series recovers
• Email headers keep the white logo visible in clients that ignore inline CSS
• Peer requests preserve uploads, downloads, query parameters, and response types
• Peer deployment actions use deployment permissions instead of Fleet configuration access
• Fleet access intersects module permissions, server-qualified user grants, and peer policy
• Operators no longer receive host shell or process-control access by default
• Certificate lists and actions are restricted to assigned deployments for non-admin users

[0.4.0-beta.4] - 2026-08-21

Fourth beta of the Albacore release, expanding operations, API discovery, and operator notifications.

Added

• System and deployment log sources, log deletion, log rules, and bounded incident triage
• Layered deployment diagnostics covering container state, Docker health, configurable application responses, proxy routing, TLS, public access, and incident lookup
• Branded nginx error responses with incident IDs in HTML and JSON
• Runtime OpenAPI discovery for routes, permissions, request bodies, and typed responses
• File and folder copy, move, archive browsing, and extraction
• Single-request deployment content pushes with optional destination cleanup
• Branded HTML email notifications with semantic status treatments, metric panels, plain-text fallbacks, and local previews

Fixed

• Deployment log reads once again honor the selected service
• Paused deployments are reported as paused instead of running
• Active deployment compose files cannot be moved out of the deployment root or deleted

[0.4.0-beta.3] - 2026-08-05

Third beta of the Albacore release, focused on the logs pillar of observability and a network-alert correctness fix.

Added

• Structured container logs: each line is parsed into a record (service, timestamp, level, message, and JSON fields) so a viewer can render logs as rows instead of a text block
• Log sources beyond container stdout: known kinds (laravel, wordpress, nginx) expose their conventional log files, and any deployment can point the viewer at a file under its own directory, confined to it; the assistant tools read these sources too

Fixed

• Network alerts no longer fire on idle containers: network metrics were a container's lifetime byte total, so a "network above X" rule crossed any threshold given enough uptime and stayed firing; the store now records a per-second rate, so alerts and charts reflect current throughput and a counter reset reads as zero
• Log level is read only from real header positions (line start, a channel.LEVEL tag, or a level= field), so a stack frame naming a class like ErrorHandler is no longer flagged as an error
• An all-logs read of a file source is bounded so a very large file cannot exhaust memory, and a stopped container's network baseline is dropped rather than retained
• The "all logs" selection returns all container output: it was passing a zero line count, which reads as no lines for a snapshot and was otherwise capped at a hundred

[0.4.0-beta.2] - 2026-08-01

Second beta of the Albacore release, continuing the deployment lifecycle and self-healing work on top of beta.1. Some Albacore items remain in progress and are not in this beta.

Added

• Deploy from a git repository: the source is fetched behind a provider interface with a registry (so upload and webhook delivery can register the same way), fetched into a temporary directory and required to contain a compose file before anything is created, with private repositories authenticating from a token held in the credential manager and scrubbed from logs
• Channel-aware updates: an opt-in prerelease channel so update can see betas, reading the full releases list with proper semver ordering, exposed over the API and surfaced as an Administration > Updates view showing the current version, the versions available to install, each version's changelog, and a one-click update that reports the restart
• Host metrics: host CPU, memory and disk collected as alertable time series using working-set memory, so a saturated machine has a signal where before there were only per-container percentages against each container's own limit
• Scheduled agents: an agent file can declare a cron schedule and a permission grant, running unattended under an actor carrying only those permissions, auto-approving the granted tools and denying the rest, failing closed so a scheduled run can never quietly do more than it was trusted to (no grant means read-only)
• Agent run history listed on its own, since each run is already a session tagged with its agent
• Agent governance policy enforcing step budgets and dry runs

Changed

• A firing alert carries a snapshot of the containers consuming the most of the resource, so the notification and the stored event name what was responsible; a rule can deliver to a chosen subset of notification targets, and can opt into restarting the offending deployment under the self-heal guardrails (managed deployments only, with a cooldown so it can't flap into a restart loop), notify-only by default
• Compose-file backups are pruned to the five most recent on each successful rewrite, so a frequently-updated deployment stops accumulating timestamped backups while one rollback copy remains

[0.4.0-beta.1] - 2026-07-29

First beta of the Albacore release: full deployment lifecycle and self-healing operations. Some Albacore items remain in progress and are not in this beta.

Added

• Observability with a native time-series UI: OTel-semconv container metrics and per-deployment serving metrics (rate, errors, average and p95) drawn from the proxy's own request record, on-disk history folded to per-minute points so the 6h and 24h ranges hold, OTLP export with a scrape endpoint, and log following
• Self-healing for FlatRun-managed deployments: unhealthy containers restart with capped retries and a cooldown so an external container can't trigger a restart loop, and the watcher reports when it gives up
• Notification system with email and webhook targets and test-send; metric thresholds alert through the configured targets
• Plugin framework: a plugin can inject UI sections, contribute settings, and expose tools the AI assistant can call
• S3-compatible remote backups to AWS, R2, B2 or MinIO on top of the always-local copy, best-effort so a remote outage never fails a backup, with object-storage secrets held in the credential manager; seeds an object-store abstraction and a MinIO template
• AI assistant tools to write a deployment file, run a quick action, start/stop/restart a deployment, and summarize a deployment's security events, each requiring deployment write access and honoring protected mode
• MCP server exposing the shared assistant tool set, and a deployment-scoped file editor whose state-changing tools pause for per-call approval even in auto-run sessions
• Agents defined as flat markdown files in .flatrun/agents/, run by the runtime through the shared tool set so permission gates, protected mode, secret redaction and the state-change approval pause all carry over; created via the assistant, the panel editor, or by dropping a file
• Listing of saved AI chat sessions, most recent first, each titled from its first message; reopening one restores its transcript and scope
• Firewall enforcement: a saved policy is translated to nftables and applied, re-applied at startup and removed when disabled, touching only FlatRun's own table and always keeping loopback, established connections and the active SSH port open
• Opt-in per-domain long browser cache for static assets, keyed off the request path's extension
• Start/stop/restart run as background jobs whose status and buffered output survive a page reload, streamed over websocket with a poll fallback and serialized per deployment
• Seeding an empty bind mount from image or running-container content, browsing a running service's files onto the host, and unmounting tagged paths
• Routing-only hostnames for externally-fronted proxies, sharing the primary domain's certificate over SNI and kept out of issuance and renewal
• Dashboards screen to create, arrange and manage panels over container and serving metrics

Changed

• Deployment status comes from a single Engine API query across all deployments instead of per-deployment docker compose ps, so listing is flat with the deployment count rather than scaling with it (list ~486ms to ~26ms at 50 deployments)
• Proxy pools connections to upstreams with real keepalive, preserving the restart rediscovery the previous per-request routing provided where the nginx image supports resolve; serving also gains HTTP/1.1 to upstreams, buffered access logging, wider gzip and larger proxy buffers
• Certificate auto-renewal defaults on and runs whenever certbot is enabled, with each certificate's own setting winning over the global default; single renew reports success only when something was reissued
• Nginx vhost generation: configurable WebSocket timeouts, conditional Connection: upgrade via a map, ssl_stapling skipped when the certificate has no OCSP responder, and the target service and port validated before a vhost is saved
• Force-recreate, no-cache rebuild and non-cached pull exposed so updated env vars and images take effect without a manual compose run
• AI assistant loop runs on the shared agents runner with per-call tool approval
• UI refresh: dark mode, design tokens, Iconify iconography, a reworked assistant and global search
• Primary-service detection prefers app/web before the first service with ports
• Example configuration ships placeholder values only

Fixed

• General-purpose HTTP clients (curl, wget, uptime monitors, webhooks) were matched as attack scanners and blocked on sight; scanner matching is now limited to self-identifying tools and ordinary clients are held to volumetric thresholds, with every auto-block recording the rule, count and paths that tripped it
• Health-watcher restart no longer holds the read lock across the docker call; notification target URLs, including SMTP passwords, are masked in API responses and stored owner-only; the host-wide auto-restart toggle is gated on settings-write rather than a deployment scope
• Additional-domain proxying routed by bare service name and intermittently served another deployment's container on a shared host; it now routes by the unique deployment name
• Compose validation resolves a relative env_file against the deployment directory on the image-set and update paths
• SaveMetadata failures during compose updates are surfaced instead of silently discarded, so compose and service.yml can't diverge
• ACME challenge locations set access_log off and log_not_found off so cleaned-up challenge 404s stop burying real errors
• Top-level --help lists update, setup and version, and help prints usage instead of starting the server

[0.3.0] - 2026-06-08

Added

• AI assistant: interactive sessions that investigate before answering, grounded in this installation's context, with inline tool-call approval, suggested actions and any OpenAI-compatible provider configured at runtime
• Server-side plans: preview and apply mutations as reviewable artifacts, with opt-in enforcement per deployment
• Optional app template for image and compose deployments: the user's image or compose content is kept while the template contributes container port, default bind mounts, pre-created directories and ownership
• Template-defined environment file generation: prefers the env example shipped inside the deployed image, falls back to template content, generates per-deployment secrets (e.g. Laravel APP_KEY) and fills in database credentials
• Interactive system terminal on a real PTY over websocket, with per-line protected-command enforcement and the global terminal disable honored
• Persisted file manager preference for showing hidden files (default on), exposed through the key-based config API
• Template mounts can declare a host path; single-file mounts (such as .env) are kept as files
• Per-service image pulling

Changed

• Mount ownership is applied recursively so nested directories belong to the container user
• Built-in template copies on disk sync automatically when the agent build changes, so upgrades take effect without a manual refresh
• Laravel template: complete storage subdirectory tree, bootstrap cache mount, env file mounted as a file
• Prompts composed by the product (log analysis, operation diagnosis) are kept out of AI session transcripts while still reaching the model
• Nginx reloads when forwarded-proxy trust settings change

Fixed

• Forwarded client IPs are only trusted when sent by configured proxies, and trusted-proxy entries are sanitized before injection into the nginx Lua layer
• e2e suite aborts on leftover root-owned state instead of failing unrelated tests

[0.2.0] - 2026-05-25

Added

• API key edit endpoint with per-deployment access levels (read, write, admin) capped by the owning user's level
• Deployment protected mode: configurable blocked actions and command rules, with an explicit enable switch
• System terminal endpoint, behind its own protected-mode configuration and a new permission
• System file manager endpoint with listing, file and directory creation, chmod, and rename
• Cheap files-info variant (?usage=false) so the UI can render the system file manager without waiting for a recursive disk-usage walk

Fixed

• API key creation by an admin key with no associated user no longer returns 401; the owning admin is resolved automatically
• apiKeyToResponse emits null instead of the zero time for unset expires_at and last_used_at

[0.1.56] - 2026-04-09

• Sync service.yml when compose expose is updated
• Add name field and fix Nextcloud compose config
• Add --non-interactive flag to certbot commands

[0.1.55] - 2026-04-07

• Migrate from mattn/go-sqlite3 to modernc.org/sqlite for pure-Go builds

[0.1.54] - 2026-04-04

• Add default server block for the nginx welcome page

[0.1.53] - 2026-04-03

• Distinguish file from directory bind mounts when materializing compose mounts