Changelog

What’s new in Falak

Every release, summarised. Tags and full commit history live on GitHub.

  1. v1.3.2

    Private networks and clearer errors

    Builders and health checks work over private networks, compose apps stop asking for variables they no longer need, and failed Cloudflare tunnel actions say why.

    • NewFALAK_BUILD_NETWORK sets the network of the BuildKit container that Docker-mode builds push from. FALAK_BUILD_NETWORK=host makes the push use the host’s network and /etc/hosts, so a builder that reaches the registry over a private address pushes the same way. Unset changes nothing.
    • ImprovedHealth checks try the server’s private address when nothing answers on its public one, for example a server behind a Cloudflare Tunnel whose provider firewall keeps ports 80/443 closed to the internet. An application error on the public address still fails the check. Off on managed (multi-tenant) control planes.
    • FixedCreating a compose app no longer fails with “The compose project needs …” for variables only read by services you moved to a Falak database or site. The check before saving now matches the preview, and the error shows even when the stack lists no variables.
    • FixedSettings → Cloudflare → Servers shows why routing a server through a tunnel, reinstalling it or removing it failed (for example a token without Account → Cloudflare Tunnel → Edit), instead of doing nothing.
  2. v1.3.1

    v1.3.0 update fixes

    Updating to v1.3 works on every install, and builds can clone git:// repositories.

    • FixedThe v1.3.0 update rolled back on some installs: the registry’s health check failed once registry token auth was on. It now checks the registry’s health endpoint, which needs no credentials.
    • FixedInstalling Docker on a server without gpg failed while checking Docker’s repository key. The agent installs gpg first.
    • ImprovedBuilds can clone repositories over git://.
  3. v1.3.0

    Hardening for shared control planes

    A security release: isolated builds per organization, scoped registry credentials, a managed (multi-tenant) mode, and signed releases.

    • NewManaged mode (FALAK_MANAGED) for control planes shared by many organizations: verified email addresses required, verification mail through Resend, domain ownership checked with a TXT record before a domain is routed, and outbound requests to user-supplied endpoints kept off private networks.
    • ImprovedShared builders keep organizations apart: native build steps run as a separate unprivileged user per organization with a clean environment, and build caches are kept per organization.
    • ImprovedThe built-in registry uses token auth: each build gets a short-lived credential for its own site’s images, and each deploy gets a one-hour pull credential for its own organization’s images only.
    • ImprovedReleases are signed: SHA256SUMS carries an Ed25519 signature, and install.sh, install-cli.sh, falak-ctl and falak update check it. Falak images are checked against the digests recorded at release.
    • ImprovedRate limits on agent endpoints and at the edge, a Content Security Policy on the panel, consoles that only accept connections from the panel’s own origin, and GitHub Actions pinned to commit SHAs.
    • UpgradeRun falak-ctl update. It creates the registry token signing keys once; existing accounts and domains are marked verified, so self-hosted installs behave as before unless FALAK_MANAGED is turned on.
  4. v1.2.4

    Panel threads on FrankenPHP 1.13

    The panel starts on FrankenPHP 1.13 again; the v1.2.3 update rolled back.

    • FixedFrankenPHP 1.13 counts worker threads separately, so the v1.2.3 panel refused to start and the update rolled back. The panel now sizes its threads for 1.13.
  5. v1.2.3

    Go 1.27.3

    Every Falak binary is built with Go 1.27.3, which fixes a certificate verification flaw in crypto/x509.

    • UpgradeThe agent, CLI, builder, relay and database helper are built with Go 1.27.3 (CVE-2026-111943). The Go functions runtime moves to golang:1.27.3-alpine.
  6. v1.2.2

    PHP2Shell patch

    Patched PHP for the panel and for FrankenPHP sites (CVE-2026-109618, CVE-2026-109619).

    • UpgradeThe panel runs FrankenPHP 1.13.2 with PHP 8.4.27. Servers get FrankenPHP 1.13.2 (glibc build), so FrankenPHP-runtime sites move to PHP 8.5.12.
  7. v1.2.1

    Pinned template catalog

    The panel image bakes the template catalog release it was tested with.

    • ImprovedThe image bakes the falak-sh/templates release pinned in templates.lock, so a fresh install starts with a known catalog before its first hourly update.
  8. v1.2.0

    Templates on their own

    The template catalog now lives in its own repository and reaches your panel within the hour, no Falak update needed.

    • NewThe 46 one-click templates moved to github.com/falak-sh/templates. Every merge there publishes a catalog release; panels fetch it hourly (sha256-checked), keep the last good copy if a download fails, and fall back to the catalog baked into the image. Settings → Templates shows the catalog version in use. FALAK_TEMPLATES_URL follows a fork, or turns updates off when empty.
    • NewTest a template like CI does: ./test <slug> in the templates repo runs Falak’s own checks inside the control-plane image (php artisan templates:render <dir>), then boots the stack with Docker and calls every public service. Templates can declare min_falak (older panels skip them) and use a Cool Icons key as their icon.
    • ImprovedAn organization’s own template keeps working when the catalog later ships the same slug: it stays editable, and deploys that name no source still pick the organization’s template.
  9. v1.1.0

    Canvas icons

    Give projects, services and groups their own icon, picked from Cool Icons right on the canvas.

    • NewIcon picker on the project canvas: click a service’s icon in its panel, a group’s icon in its frame, or the project icon at the top right, then search Cool Icons (coolicon.falak.sh), pick an icon and a style, or go back to the default. Icons show on cards, panels, the project switcher and the Projects dashboard.
    • NewIcons are UI only: falak.yaml never reads or sets them. Set a service’s icon over the API with PATCH /api/v1/projects/{project}/environments/{environment}/services/{service}; FALAK_ICONS_URL points the picker at a self-hosted copy, or turns it off when empty.
  10. v1.0.1

    Caddy v2.11.7

    Rate limits and DNS-01 certificates work again on servers whose Caddy was upgraded to v2.11.7.

    • FixedServers whose apt Caddy moved to v2.11.7 kept their stock Caddy: Falak’s Caddy build (v2.11.4) lacked modules the newer Caddy has, so the agent refused to switch, and rate limits and DNS-01 certificates fell back. Falak’s Caddy build is now v2.11.7.
  11. v1.0.0

    Falak 1.0

    Falak moves to the falak-sh GitHub organization, connects one GitHub App per account or organization, and a DNS-01 domain no longer breaks a server’s routes.

    • NewOne GitHub App per GitHub account or organization: add an app for each owner in Settings → Source control (+ Add GitHub App) and install it on that owner’s repositories.
    • ImprovedFalak now lives at github.com/falak-sh/falak, and its images at ghcr.io/falak-sh. falak-ctl update moves an install over by itself once the new registry serves the version; forks and custom registries are left alone.
    • FixedA domain with a DNS-01 certificate (tls: dns) made every route update of its server fail on stock Caddy, which has no DNS provider module. The agent now installs Falak’s Caddy build for DNS-01 too, or falls back to an HTTP-01 certificate and applies everything else.
    • UpgradeUpdate the agents after the control plane (Servers → Update all agents): DNS-01 domains need agent v1.0.0.
  12. v0.25.1

    Large releases

    Builds whose release is over 64 MB deploy again.

    • FixedA build whose release was over 64 MB (a static site with videos, for example) failed with "upload artifact: HTTP 413": the control plane applied its form-size limit to the builder’s streamed upload. Uploads now only stop at the artifact limit (FALAK_ARTIFACTS_MAX_BYTES, 4 GiB by default).
  13. v0.25.0

    Tailscale

    Put servers on your tailnet, make sites reachable from your own devices only, and close the panel to everyone else.

    • NewTailscale: save a reusable auth key in Settings → Tailscale and join servers from their Network tab. Each server advertises its own public IPs as routes, so devices that accept routes reach it through the tunnel; DNS, certificates and agents keep using the public path.
    • NewTailscale only, a switch in a site’s Networking tab: only your tailnet devices (plus the Allow only entries and Falak’s own checks) reach the site, everyone else gets 403. It needs every server of the site on the tailnet with its routes approved, and domains that are DNS only (not proxied by Cloudflare).
    • Newfalak-ctl panel-access tailnet | public | status closes the panel to browsers outside the tailnet. Agents, webhooks, the token API, consoles and public status pages stay public; falak-ctl panel-access public over SSH switches it back.
    • UpgradeJoining needs agent v0.25 or newer; servers on older agents are asked to upgrade first.
  14. v0.24.0

    Contabo

    Create servers on Contabo from Falak: paste one line from the Contabo API page and Falak does the rest.

    • NewContabo as a cloud provider: paste the cntb set-credentials line (or the four API values) as one block; Falak picks out the Client ID, Client Secret, API User and API Password, verifies them and names anything missing. Contabo bills monthly: deleting a server cancels it at the end of the paid month.
    • ImprovedDocumentation in the panel opens falak.sh/docs.
    • ImprovedServers on slower providers get up to 20 minutes to receive their IP address.
  15. v0.23.0

    One machine is enough

    Deploy and build on the machine that runs Falak without risking the panel: a protected share keeps it up when a site or a build runs away. Update Falak from the panel, and move sites to another server when you outgrow one box.

    • NewThis machine in Add server: the control-plane host becomes a regular server. Its front door sorts connections by name, so sites get their own domains and TLS next to the panel (HTTP/2; no HTTP/3 for sites there).
    • NewHard caps: the panel keeps a reserved share (1.5 GB by default) with top priority; sites, builds and everything Falak deploys there share the rest, capped in memory, swap, CPU, tasks and I/O. A runaway build or site is stopped, never the panel.
    • NewBuild on the site's own server (build mode on-server): builds run in a capped builder there, one at a time, with a disk guard that pauses builds before the disk fills.
    • NewUpdate Falak from the panel: owners and admins see new releases with their notes and update with one click — a backup first, a live log, and an automatic rollback if the new version fails.
    • NewMove a site to another server, or move all sites off one: databases and volumes are carried over, managed DNS follows, and the old server keeps serving until the new one takes over.
    • ImprovedSites on the control-plane machine can't reach the panel's networks, volumes or containers, and can't use host namespaces, host mounts, devices or extra capabilities.
    • Upgradefalak-ctl update --version v0.23.0, then sudo falak-ctl up once to install the host units (the panel shows this hint), then update the agents.
  16. v0.22.0

    AI templates

    Twelve new one-click templates for AI: gateways that put every model behind one key, autonomous agents, chat and RAG apps, LLM platforms and private web search, each secured with generated credentials.

    • NewAI gateways: OmniRoute, CLIProxyAPI with its management console, 9router and LiteLLM (with Postgres). Each answers on an OpenAI-compatible /v1 that needs a key.
    • NewAgents: Hermes Agent with Nous Research's dashboard behind a password, and OpenClaw with its Control UI (gateway token plus device pairing).
    • NewApps and platforms: AnythingLLM, Flowise, LibreChat, Dify and Langfuse; plus SearXNG, a private search engine your agents can query as JSON.
    • ImprovedA compose service labelled falak.environment_network: none is kept off the environment network where your databases answer; Dify's code sandbox uses it.
    • Upgradefalak-ctl update --version v0.22.0, then update the agents.
  17. v0.21.0

    Sandbox snapshots and forks

    Freeze a sandbox with its memory, running processes and files, then start new sandboxes from that moment: one to undo, or up to ten to try several approaches at once.

    • NewSnapshots: falak sandbox snapshot captures memory, processes and files; processes keep running where they were in every sandbox made from it. Named snapshots work as templates (falak sandbox create --from node-ready).
    • NewFork: falak sandbox fork <snapshot> -n 3 starts up to ten independent sandboxes from one snapshot, each with its own address.
    • NewPause and resume, and idle pause: a sandbox nobody uses for a while is paused and wakes on the next command, file call or console.
    • NewDatabase copy: falak sandbox db <sandbox> <instance>/<database> streams a dump of a database on the same server into the sandbox; the sandbox gets the data, never access to the database.
    • NewMCP: sandbox_snapshot, sandbox_fork, sandbox_pause, sandbox_resume and sandbox_list_snapshots, so agents can snapshot before a risky step and try several fixes in parallel.
    • Upgradefalak-ctl update --version v0.21.0 (runs migrations), then update the agents and the CLI (falak update).
  18. v0.20.0

    Sandboxes

    Disposable Linux machines on your own servers for AI agents and untrusted code, isolated with gVisor: create, run commands, move files, open a shell, and they die on their own when their time is up.

    • NewSandboxes: falak sandbox create|ls|show|exec|cp|ssh|extend|kill, the API under /api/v1/sandboxes, and a Sandboxes page in the dashboard with a console, the exec log and a file browser.
    • Newfalak mcp: an MCP server so Claude Code and other agents can create sandboxes, run commands and read or write files (claude mcp add falak -- falak mcp).
    • NewNetwork modes: none, internet (no private networks, cloud metadata, the server or other sandboxes) and allowlist (only the hosts you list, through a proxy).
    • NewLimits and quotas: CPU, memory and disk per sandbox; running sandboxes, total memory and total disk per organization; a time limit enforced by both the control plane and the agent.
    • NewPermissions sandboxes.view, sandboxes.use and sandboxes.manage, also as token abilities, so an agent can get a token that can only drive sandboxes. Servers can opt out of running sandboxes.
    • ImprovedCommand results that hold sandbox output or file contents are no longer kept by the control plane or copied into its queue.
    • Upgradefalak-ctl update --version v0.20.0 (runs migrations), then update the agents (gVisor is installed on a server the first time it runs a sandbox) and the CLI (falak update).
  19. v0.19.2

    Config export round trips

    An exported falak.yaml now validates and plans to no changes, the live deployment never reads Removed, and falak --help fits on one screen.

    • FixedDeployments: the live deployment no longer shows a Removed badge right after it finishes, in the panel and the Deployments tab.
    • Fixedfalak config export: -e defaults to production, so the command works without flags.
    • FixedConfig as code: services, schedules and daemons made in the dashboard export under names that validate (Nightly report becomes Nightly-report), and planning the exported file maps every unit back to the same service: export, validate, plan shows nothing to change.
    • FixedConfig as code: a schedule or daemon the file manages keeps its key; a dashboard item named like it gets -2 instead of taking it over.
    • Improvedfalak --help: a compact command list grouped by area.
    • Upgradefalak-ctl update --version v0.19.2 (no migration), then update the agents and the CLI (falak update).
  20. v0.19.1

    Readable icons and named network charts

    Brand icons stay readable in both themes, network charts name their peers, chart labels never wrap, and Talks to counts the short-lived connections of sites running on the host.

    • FixedTemplate and service icons: dark brand colours (Odoo) are lightened in dark mode and pale ones darkened in light mode, so every logo stays readable.
    • FixedNetwork: the TCP retransmits and per-peer charts show the database, server or site name instead of an internal id.
    • FixedCharts: y-axis labels such as 954 MB or 6,000 ms no longer wrap or clip.
    • ImprovedTalks to: with eBPF on, connections opened by native (PHP-FPM) sites show up even when no socket was open at the snapshot.
    • Upgradefalak-ctl update --version v0.19.1 (no migration), then update the agents.
  21. v0.19.0

    A new canvas, function drafts and the AI skill

    The canvas gets a space backdrop and a keyboard-first Create picker, functions start as drafts, volumes open beside their service, Odoo and ERPNext join the catalog, and AI agents get a Falak skill.

    • NewCanvas: a space backdrop (moons, stars and a slow nebula, still under reduced motion, a pale ink variant in light mode) and a centered Create picker driven by one keyboard list: type to filter, arrows, Home/End, Enter, and Esc to go back one level.
    • NewFunctions created on the canvas start as drafts: the panel opens on the Code tab and nothing runs until you press Deploy, with a Not deployed yet notice. The API still deploys right away unless told otherwise.
    • NewA volume chip on a canvas card opens the volume (Overview, Backups, Files) over its service’s panel, with the canvas still in view.
    • NewTemplates: Odoo 19 with PostgreSQL 17, and ERPNext v16 (the frappe_docker stack, 11 services), in a new Business category.
    • NewThe Falak skill for AI agents (Claude Code, Codex and other SKILL.md tools): falak skill install [--user | --project | --dir DIR], with references for the CLI, falak.yaml, the API, workflows and safety, and least-privilege token advice.
    • Upgradefalak-ctl update --version v0.19.0 (no migration), then falak update for the CLI and the skill command.
  22. v0.18.1

    Fixes from the UI walk-through

    A Network tab on servers, a Network area with its own tabs, charts scaled to usage, safer Block buttons and short-lived connections in Talks to.

    • ImprovedServers get a Network tab: connections, listeners, talks-to, TCP & DNS with the eBPF switch, then the network tools and packet capture. The Private network tab is back to WireGuard only.
    • ImprovedNetwork has its own tabs: Private networks, Blocked IPs and Rate limits. /uptime redirects to Observability → Uptime.
    • ImprovedCharts scale to usage: limits and capacity are drawn only when they are within 2× of it, otherwise named in the legend with their value.
    • ImprovedNo Block button on addresses that must stay reachable: they are labelled instead (Control plane, Your server, Private network, Team member, Cloudflare, Private range, Reserved, Your address).
    • ImprovedWhile the eBPF probe runs, short-lived outbound connections (such as a database connection per request) show up in Talks to (agent feature net.ebpf.connects).
    • FixedWorkload memory counts shared memory (PostgreSQL shared_buffers, tmpfs) as used, not cache; SSH hits show on the rule that allows SSH; short mixed ids in routes become :id; the site Observability tab falls back to the edge’s request counts; consoles get their fitted size when they open; the eBPF probe’s burst check skips its first pass and a stale over-budget stop is cleared.
    • Fixedfalak logs --level accepts warning as warn instead of failing.
    • Upgradefalak-ctl update --version v0.18.1 (no migration), then update every agent.
  23. v0.18.0

    eBPF network insight and packet capture

    The last phase of the observability plan: an opt-in kernel probe for TCP health, DNS and run-queue delay, and short packet captures you download as a .pcap.

    • NeweBPF network insight, per server and off by default (Admin only): a TCP & DNS section on the server, site and database Network tabs with retransmits, resets, connect latency and RTT per workload and peer, DNS queries, latency, failures and top names, and run-queue delay. Alerts on DNS failure and retransmit spikes.
    • NewPacket capture on demand (Admin only): up to 60 s on a server interface or inside a site’s or database’s container, filtered by host, port, protocol and direction, downloaded as a .pcap. Packet headers by default, full packets only when you ask. falak net capture does the same from the CLI.
    • ImprovedThe probe only reads, needs Linux 5.8+ with BTF, is removed by the kernel if the agent stops, and switches itself off above 1% of CPU until an admin re-enables it.
    • ImprovedCaptures are confirmed, audited, rate-limited (3 at once, 10 an hour per organization), encrypted on the control plane and deleted after 24 hours.
    • Upgradefalak-ctl update --version v0.18.0 (migrations only add tables), then update every agent (features net.ebpf and net.capture). tcpdump is installed on the first capture if missing.
  24. v0.17.2

    Block IPs, SSH brute-force protection and edge rate limits

    Act on what the Network and HTTP tabs show: block addresses, catch SSH brute force, see firewall hits and rate-limit paths at the edge.

    • NewBlock an IP or range from Network → Blocked IPs or from any connection, client or threat row: one server or the whole organization, with an optional expiry and a reason, audited. Blocks drop traffic before anything else, published container ports included, in their own nftables table.
    • NewSSH brute-force detection opens security Issues. Auto-ban is optional and off by default, set per server (N failures in M minutes bans for T).
    • NewFirewall rule counters: packets and bytes per rule on the Firewall tab, and how often each block was hit.
    • NewPath rate limits at the edge (Network → Rate limits), e.g. 10 requests a minute per client IP on /login. The first rule on a server installs Falak’s Caddy build, checksum-verified and rolled back if the edge doesn’t come back. Not on FrankenPHP edges.
    • ImprovedLock-out guards: blocks and bans never cover your own IP, the control plane, your servers and private networks, addresses your members used in the last 30 days, a server’s own addresses, resolvers or gateway.
    • Upgradefalak-ctl update --version v0.17.2, then update every agent (feature net.protection; rate limits also need the edge feature).
  25. v0.17.1

    Uptime, status pages, network tools and Cloudflare analytics

    Know when a site is down, publish a status page, debug the network from the panel and see what Cloudflare served.

    • NewUptime monitoring: every site domain is checked from the control plane once a minute (status, response time, certificate chain and expiry). Add your own HTTP(S), TCP and TLS monitors every 30 s to 10 min. Incidents, uptime Issues and alerts (down, recovered, TLS problem, slow).
    • NewPublic status pages at /status/<slug>, opt-in for admins, showing only the monitors and names you choose.
    • NewNetwork tools: ping, traceroute, DNS lookup, HTTP timing and TCP tests from a server or inside a site’s or database’s container, streamed live and audited. falak net from the CLI. Admins only.
    • NewCloudflare analytics for a site’s proxied hostnames: requests, cached vs uncached, bandwidth, threats and countries. The token needs Analytics: Read.
    • ImprovedUptime checks never reach private, loopback, link-local or metadata addresses, and are recorded as unknown when the control plane itself is offline.
    • Upgradefalak-ctl update --version v0.17.1. Network tools need agent feature net.tools; uptime and Cloudflare analytics run in the control plane.
  26. v0.17.0

    Network connections and HTTP analytics

    What reaches your sites, what they talk to and what’s exposed, all read-only.

    • NewA Network tab on servers, sites, containers and databases: connections by state, listening ports and their owners, the top peers each workload talks to, long-lived connections, conntrack usage, and WireGuard peer health and latency.
    • NewExposed-ports audit: a public listener the firewall doesn’t cover becomes a security Issue.
    • NewAn HTTP tab per site from the edge’s access logs, no APM needed: requests, status classes, bytes, top and slowest routes, clients, user agents, HTTP and TLS versions, countries and networks (DB-IP Lite), and a Threats view of scanners.
    • NewSlow endpoints become performance Issues; bandwidth per site per month with an optional alert; opt-in X-Request-Id stamping at the edge.
    • NewAlerts: conntrack near full, WireGuard peer stale, latency spike, 5xx spike, slow endpoint, bandwidth threshold, scanner surge.
    • Upgradefalak-ctl update --version v0.17.0, then update every agent (features net.connections and edge.request_id).
  27. v0.16.2

    PostgreSQL top queries reach the panel

    A fix to the falak-db helper in the PostgreSQL image.

    • FixedPostgreSQL top queries, largest tables and unused indexes failed with “pg_stat_statements: unexpected output”; the helper now parses the whole result.
    • Upgradefalak-ctl update --version v0.16.2, update agents, then apply the new image from each PostgreSQL instance’s page (a brief restart).
  28. v0.16.1

    Log exceptions the APM never saw reach Insights

    Exceptions from Horizon, queue supervisors, artisan commands and Log::error() no longer get dropped for sites that use the APM package.

    • FixedA Laravel log exception is now dropped only when the APM reported the same exception (same class and first in-app frame) within a minute, also for apps whose timezone isn’t UTC.
    • UpgradeAgents only. falak-ctl update --version v0.16.1 keeps versions aligned; no migrations.
  29. v0.16.0

    Database metrics, slow queries and insights

    Managed PostgreSQL, MySQL, MariaDB, Redis and Valkey instances get Metrics, Logs and Insights tabs.

    • NewEngine metrics every minute: throughput, cache hit rate, connections by state, locks and deadlocks, slow queries, replication lag and the largest databases and tables; Redis and Valkey memory against maxmemory, evictions, clients and keys.
    • NewSlow-query and engine logs per instance, and top queries hour by hour for 30 days. Query text is normalized; raw PostgreSQL SQL and bind parameters are filtered out of shipped logs.
    • NewDatabase Issues (slow query shapes, engine errors) and recommendations (missing and unused indexes, low cache hit rate, Redis near maxmemory, connections near the limit).
    • NewAlerts: slow-query spike, replication lag, deadlocks, memory near the limit.
    • ImprovedUpgrading never restarts a database. Top queries for existing PostgreSQL and MariaDB instances need a restart you choose (Enable query insights).
    • Upgradefalak-ctl update --version v0.16.0 (one migration), update agents (db.observability). Engine figures need the v0.16.0 database images; updating an image recreates the container.
  30. v0.15.0

    Logs from inside containers, PHP-FPM logs

    Log files written inside containers, PHP-FPM error and slow logs, and Laravel exceptions as Issues without the APM package.

    • NewSite → Logs → Sources: up to 10 paths or globs inside a Docker or compose site’s containers, read from the host with no docker exec, with an optional start-of-record pattern.
    • NewPHP-FPM error logs and a slow-request log with stack traces (5 s by default for new sites, off for existing ones until you set it).
    • NewLaravel ERROR and CRITICAL entries with a stack trace open or update Issues, without the APM package.
    • NewA source filter in the Logs tab: output, app files, container files, PHP-FPM, slow requests, access.
    • Upgradefalak-ctl update --version v0.15.0 (one migration), then update agents (logs.container_files, runtime.fpm.slowlog). No PHP-FPM pool is reloaded by the upgrade.
  31. v0.14.1

    Usage of sites that share a system user

    Native sites that run as the shared falak user get their own CPU, memory, disk and process charts.

    • FixedEach process is assigned to its site: PHP-FPM workers by pool, workers, Horizon, schedules and consoles by the site folder they run in.
    • Upgradefalak-ctl update --version v0.14.1, then update agents (telemetry.workloads.shared).
  32. v0.14.0

    Metrics per workload

    A site’s Metrics tab shows what the site itself uses, against its server and its limits, instead of the whole server.

    • NewContainer sites: CPU, memory, disk I/O, network and processes from the container’s cgroup, stacked on the rest of the server, with limit lines, CPU throttling and OOM kills. Compose sites get a per-service table.
    • NewNative sites: the same charts from the site’s slice or its own system user.
    • NewRequests, 5xx and p95 latency from the edge’s access logs when the app sends no traces.
    • NewA server picker for sites on several servers, and the Grafana dashboard “Falak / Workloads”.
    • Upgradefalak-ctl update --version v0.14.0, then update agents (telemetry.workloads).
  33. v0.13.0

    Consoles for sites, containers and databases

    Open a shell where your app runs, from the panel or the CLI, without SSH.

    • NewSite consoles: a Docker site opens its container, a compose site the service you pick, a native site a shell as the site user in its current release with its variables. falak console <site>.
    • NewDatabase consoles (falak console --db) and server shells from the CLI (falak console --host) with no SSH key or port 22.
    • NewThe terminal relay: a small Go service that carries consoles and terminals, so typing costs one round trip. The page falls back to the previous path if the relay can’t be reached.
    • ImprovedSessions stay audited, recorded and shareable; removing access disconnects the clients that lost it.
    • Upgradefalak-ctl update --version v0.13.0 adds the relay service (one migration). Update agents (terminal.v2) and the CLI (falak update).
  34. v0.12.3

    Composer on builders, first deploys of new units

    Fixes from deploying a Laravel app and a static site natively through falak.yaml.

    • FixedComposer builds failed on builders started by systemd (no HOME); builds now get a COMPOSER_HOME in the builder’s cache.
    • Fixedfalak.yaml: a new native unit gets its domains before its first deployment, so its first health check has a route through the edge.
    • Fixedfalak.yaml: switching a unit between docker and a native runtime is refused in the plan, with a hint, instead of failing while applying.
    • Upgradefalak-ctl update --version v0.12.3 (no migration), then update builders’ agents.
  35. v0.12.2

    Add PHP and Node to an existing server

    A server created without PHP or Node.js can get them later, and builders can have the PHP CLI for Composer builds.

    • Newfalak servers stack <server> --php 8.4 --node 22 (or POST /api/v1/servers/{server}/stack) records the runtimes and re-provisions the server. Add-only.
    • NewBuilders can have the PHP CLI (--php-runtime cli); a builder never gets a web server. Builders without Composer use a checksum-verified Composer 2.10.3.
    • FixedA config run no longer stays “applying” forever when a deployment it queued already failed.
    • Upgradefalak-ctl update --version v0.12.2 (no migration), update builders’ agents, then falak update.
  36. v0.12.1

    Edge fixes for servers set up before v0.10

    Two agent fixes that kept routing changes from reaching the edge (Caddy) of servers provisioned before v0.10.

    • FixedThe agent now updates an older falak-edge.service when it starts, so /run/falak-edge (the admin socket’s directory) exists and Caddy accepts new configs again.
    • FixedAdmin calls over Caddy’s unix socket no longer send an Origin header, which Caddy refused with 403 (switching upstreams on a deploy, loading a config).
    • UpgradeUpdate every agent. On a server whose edge shows Failed, click Re-apply after the update.
  37. v0.12.0

    Log in through the browser

    falak login opens the panel: sign in as usual, check the code, approve, and the CLI gets its own token. No token to copy by hand.

    • NewBrowser login for the CLI (RFC 8628 device flow): approving needs a fresh password and 2FA check, shows the asking machine and address, and lets you pick the organization. --no-browser prints the link for SSH sessions.
    • NewCLI tokens are named “CLI — your-host”, carry your role’s permissions, and expire after 90 days without use and a year after the login at most. falak logout revokes them on the server.
    • NewAPI: POST /api/v1/cli/login, POST /api/v1/cli/login/token and DELETE /api/v1/me/token.
    • FixedThe falak init --update tests that v0.11.1 overwrote are back, and a flaky logs --follow test is fixed.
    • Upgradefalak-ctl update --version v0.12.0 (one migration), then falak update for the CLI. Tokens from Settings, --token-stdin and FALAK_TOKEN work as before.
  38. v0.11.1

    falak secret and falak update

    Two CLI commands: set the secrets a falak.yaml references without leaving the terminal, and update the CLI in place.

    • Newfalak secret set | list | delete for project and environment secrets. The value comes from a hidden prompt or stdin, never an argument; setting an existing secret adds a version; --from-env imports a .env file; list never shows values.
    • Newfalak update replaces the CLI binary with the latest release, verified against the release SHA256SUMS; --check only reports, --version pins a release.
    • Improvedfalak init now points at falak secret set for the secrets the file references.
    • UpgradeCLI only: the control plane and agents are unchanged from v0.11.0. Update the CLI with the install line once; later, falak update.
  39. v0.11.0

    Config as code

    Describe a whole project in one falak.yaml at the root of its repository: apps, compose stacks, functions, databases, Redis/Valkey, volumes, backups, domains, workers, schedules, limits and previews, per environment. A push validates the file, plans the changes, applies them and deploys only what changed.

    • Newfalak init writes a first falak.yaml from the code (Laravel, Symfony, Next, Nuxt, Vite, Astro, Bun, Deno, static sites, Dockerfiles, compose files, monorepos), each value with the reason; secret-looking values become ${{ secrets.NAME }} references. falak init --update proposes only what the code added.
    • NewPlan before you apply: falak validate (offline, with line and column), falak plan, falak apply --wait; pull requests get the plan as a comment and a falak/config commit status. Env values are masked in every plan.
    • NewApprovals for risky changes (database major upgrades, public databases, removing production domains, changing production servers, lowering backups, removing data units), bound to the exact plan shown and expiring after 7 days.
    • NewThe file is the source of truth: fields it manages are read-only in the dashboard, removed resources are detached rather than deleted, and a push never deletes data. Drift is checked every 30 minutes; rolling back restores the config saved on the release.
    • NewNew CLI commands and API endpoints: runs, approve, reject, drift, config export, config source, config unmanaged, db list, db delete; a JSON Schema at falak.sh/schema/v1.json for editor completion.
    • ImprovedA failed health check shows the container state and its last 50 output lines (secrets masked); validation reports a missing PHP version or runtime before anything runs.
    • UpgradeUpdate every agent after the control plane. API tokens created with a list of abilities need the config.* abilities for the new commands. Nothing changes for a project until its config source is enabled.
  40. v0.10.0

    Production-ready

    The release that makes Falak safe to trust with production: a secret store, databases as containers with point-in-time recovery, encrypted backups with restore drills, resource limits, rollback after a release goes live, alerts everywhere, a server security baseline, disaster recovery and preview environments.

    • NewSecret store: organization, project, environment and service secrets with versions, an access log and re-authenticated reveal, referenced as ${{ secrets.NAME }}; external providers (Vault / OpenBao, AWS Secrets Manager and SSM, 1Password Connect, Doppler, Infisical, HTTPS webhook). On servers, secrets live in memory only and are masked in every log.
    • NewDatabases as containers: PostgreSQL 15–18, MySQL 8.0/8.4, MariaDB 10.11/11.4, Redis 7.4/8 and Valkey 8.1 from signed, pinned Falak images, one container per database with its own volume, memory limit, TLS and private network.
    • NewPoint-in-time recovery for PostgreSQL, MySQL and MariaDB: changes ship every minute, and a restore to any second lands in a new read-only copy you can swap in, keep or discard.
    • NewEncrypted backups (Falak-held or your own age key), verified before every restore, with automatic restore drills and falak-restore for offline restores.
    • NewVolumes: sized volumes you can grow online, backups, clone, move between servers and a file browser.
    • NewResource limits for every service (memory, CPU, processes, restarts, logs), with out-of-memory and restart-loop alerts and a capacity view per server.
    • NewRollback after a release goes live: an opt-in watch window rolls back when health checks fail, the service crash-loops or its error rate jumps.
    • NewAlerts everywhere — disk with a 48-hour forecast, memory, CPU, expiring certificates, missed backups, recovery lag, database connections, secrets and security — with a default rule pack for every organization.
    • NewServer security baseline: a daily audit, a score and a Production ready badge, with one-click fixes that can be undone for 7 days.
    • NewDisaster recovery: scheduled, encrypted and authenticated control-plane backups to S3, restore onto a fresh host, monthly drills, and a wizard that rebuilds a lost server.
    • NewPreview environments for every pull request on GitHub, GitLab and Bitbucket, on your own preview domain; pull requests from forks need a member's approval and never get secrets.
    • UpgradeBreaking: install the v0.10.0 falak-ctl before updating, update every agent right after the control plane, and sign in again. Databases installed directly on servers are no longer managed — see the install guide's upgrade notes.
  41. v0.9.0

    Redis backups and Ubuntu 26.04

    Redis and Valkey get backups and restores, and Ubuntu 26.04 is supported.

    • NewRedis and Valkey backups and restores, with version checks, AOF restores and free-space checks before restoring.
    • NewUbuntu 26.04 servers.
    • ImprovedDatabase references are never resolved to public addresses; unresolved connections are flagged on the canvas.
    • NewDelete sites through the API (DELETE /api/v1/sites/{site}).
    • FixedImage pulls retry on temporary registry errors.
  42. v0.8.2

    Aurora teal

    The panel moves to the Aurora teal palette, in light and dark, with contrast checked throughout.

    • ImprovedAurora teal across the panel, favicons and e-mails; success stays green and info moves to blue so statuses never blur into the accent.
  43. v0.8.1

    The Falak mark

    The panel gets the real Falak logo, the orbit sphere, everywhere it shows a brand.

    • NewThe Falak mark in the top bar, sign-in pages, favicons and e-mails, sharp at every size and in both themes.
  44. v0.8.0

    Falak

    The product has a new name — Falak (falak.sh) — renamed from its working title with no automatic migration from earlier installs, which simply reinstall.

    • NewRenamed from its working title to Falak (falak.sh): binaries falak, falak-agent, falak-ctl, FALAK_* env vars, /opt/falak, images ghcr.io/falak-sh/falak-*.
    • NewFalak Cloud, a hosted control plane, at https://cloud.falak.sh.
    • NewOperator-owned extra sites on the control-plane host from /opt/falak/edge, mounted read-only and imported after the generated sites.
    • ImprovedThe installer moves to https://falak.sh/install.sh.
    • UpgradeAPM for Laravel supports Laravel 12 and 13; Laravel 11 is dropped (out of security support).
  45. v0.7.1

    Redis, reachable

    A Redis or Valkey instance is now reachable from containers on its own server and from other servers over a private network, and a Compose stack’s `redis` or `valkey` service can become a managed instance.

    • NewNetwork access for Redis and Valkey: containers over the Docker bridge, other servers of the environment over a private network (WireGuard, then the provider’s), never public.
    • NewA Compose stack’s redis or valkey service can run as a managed Redis/Valkey instance, with its references rewritten automatically.
    • ImprovedThe Redis panel shows who can connect and from where, with the reason when a site can’t.
    • FixedInstances come back up on their Docker or WireGuard address after a reboot instead of racing those services at boot.
  46. v0.7.0

    Redis and Valkey

    Create isolated Redis or Valkey instances next to your PostgreSQL, MySQL and MariaDB databases, connect with `REDIS_*` variables, and tune memory, eviction and persistence from the panel.

    • NewRedis and Valkey as database engines: one instance per service, with its own port, password, memory limit, eviction policy and persistence.
    • NewRotate an instance’s password and switch its persistence mode without losing data.
    • ImprovedA stock Redis the server already runs is left untouched — Falak’s instances can’t read or overwrite it, and it can’t touch theirs.
  47. v0.6.0

    Machine check

    Before provisioning a server, Falak now checks what’s already installed — Docker, nginx, a vendor PostgreSQL — and reuses, completes or stops with a fix instead of overwriting it.

    • NewMachine check runs after the agent enrolls and on every Re-provision; it only reads the machine until provisioning starts.
    • NewNeeds attention status and a Machine check panel on the server page, with Re-check and Provision actions.
    • ImprovedFalak never replaces or removes software it didn’t install.
  48. v0.5.2

    Smoother reinstalls

    Reinstalling a server replaces its agent identity cleanly, the installer runs preflight checks first, and provisioning recovers from a broken apt/PPA or a missing PHP package.

    • NewInstaller preflight: checks the OS, that the panel and agent API are reachable, and the clock, before replacing an earlier install.
    • FixedReinstalling a server replaces its agent identity instead of keeping the deleted server’s.
    • FixedProvisioning recovers from a PPA without a release and falls back to the distribution’s PHP.
    • FixedPrivate IPv4 is read from real interfaces only (the default route’s first).
  49. v0.5.1

    Compose services, in order

    Split a Compose service out into its own site at creation time, and Falak deploys everything in the order it needs.

    • NewSplit a Compose service out into its own site at creation, not just after the fact.
    • ImprovedAutomatic deploy order: split-out sites bootstrap before the stack that depends on them.
    • FixedSplit-out sites inherit their env file’s keys, and database names are freed once they’re no longer used.
  50. v0.5.0

    Compose apps from a repository

    Deploy a Compose stack straight from your repository — several files, profiles, per-service choices — with edge routing for every public service and a built-in image registry.

    • NewDocker Compose apps from a git repository: several files, profiles, include/extends, and Falak-specific adjustments.
    • NewRun a Compose service as its own Falak site or as a Falak database, with its own domains and health checks.
    • NewBuilt-in image registry for production installs, with weekly garbage collection.
    • NewEdge routing — domains, Cloudflare rules, health checks — for every public Compose service.
    • ImprovedContainers can reach database engines on the same server over the Docker bridge.
  51. v0.4.4

    Go, multi-file functions, and cleanup

    Functions can span several files and run on Go, telemetry never leaks secrets in URLs, and a batch of housekeeping fixes landed alongside.

    • NewGo runtime for functions (net/http Handler, Scheduled, static builds) with 10 starters.
    • NewMulti-file functions: a file tree and per-file diffs.
    • ImprovedFunction telemetry never includes secrets in URLs, across every runtime.
    • NewFALAK_REGISTRATION=open|invite|closed restricts sign-up; invite-only requires the invitation link.
    • Fixedfalak-ctl removes old images after a successful update (prune-images command).
  52. v0.4.3

    Lock down and test your functions

    Restrict who can call a function, mount it at a path on another site, send it a test request, and manage it all from the `falak fn` CLI.

    • NewAccess control for functions: API keys and IP allowlists.
    • NewMount a function at a path on another site (edge mounts).
    • NewTest request panel, and observability charts.
    • Newfalak fn CLI: list, pull, deploy, versions, rollback, run, logs, invoke.
  53. v0.4.2

    More languages for Functions

    Write functions in Node.js, Deno or Python, not just Bun, each with 10 ready-to-use starters.

    • NewNode.js 24, Deno 2 and Python (ASGI, uv) runtimes for functions.
    • New10 starters per language; the TypeScript ones are shared across Bun, Node and Deno.
    • ImprovedA shared tracer, entry loader and scheduled runner behind every JavaScript runtime.
  54. v0.4.1

    Functions on a schedule

    Give a function a cron schedule, trigger a run on demand, and see its run history — including the runs it missed.

    • NewSchedules (cron) for functions, with Run now and run history.
    • NewMissed-run detection, and an Insights issue that names the schedule when a run fails.
    • ImprovedA run that hits its timeout exits cleanly and keeps its logs.
  55. v0.4.0

    Cloud Functions

    Write a function in Falak’s own editor and deploy it in seconds — Bun + Hono to start, scaling to zero between requests with built-in observability.

    • NewFunctions: write code in the browser, deploy in seconds, no repository or Dockerfile.
    • NewBun + Hono runtime with scale to zero, versions and rollback.
    • NewBuilt-in observability: request spans, exceptions, outgoing fetches, cold starts, live status.
  56. v0.3.2

    Cache, purge, lock down

    Set cache behavior per domain, purge automatically after each deploy, flip on Under Attack mode, and lock your origin down to Cloudflare or a tunnel.

    • NewCache modes per domain, with an automatic purge after every deploy.
    • NewUnder Attack mode and origin lock-down (Cloudflare-only, or closed behind a tunnel) from the panel.
    • ImprovedLock-down holds against all-port firewall rules and port ranges; a lost tunnel is detected.
  57. v0.3.1

    No open ports, just a tunnel

    Route a server’s traffic through Cloudflare Tunnel instead of opening inbound ports — Falak installs and runs `cloudflared` for you.

    • NewCloudflare Tunnel as a server’s ingress: cloudflared installed and managed by Falak.
    • NewDomains route through the tunnel once it’s up; no open inbound ports needed.
    • ImprovedDNS only moves to the tunnel after cloudflared is confirmed running.
  58. v0.3.0

    Cloudflare, connected

    Connect a Cloudflare token and Falak keeps DNS in sync with your services automatically, issues certificates through the proxy, and sees visitors’ real IPs.

    • NewConnect a Cloudflare API token; Falak finds your managed zones.
    • NewDNS records created, updated and removed automatically for Falak’s own services, with names generated under your zone.
    • NewHTTP-01 certificates behind Cloudflare’s proxy (orange cloud) — no DNS plugin needed on servers.
    • ImprovedReal visitor IPs are trusted only from Cloudflare’s published ranges.
  59. v0.2.8

    Update your fleet, one click

    Update agents across selected servers or the whole fleet from the Servers list, and pick from 14 more one-click templates.

    • NewUpdate one agent, a selection, or every server’s agent from the Servers list.
    • New14 new templates: Ollama + Open WebUI, Langflow, Qdrant, changedetection.io, ntfy, Nextcloud, Paperless-ngx, BookStack, Wiki.js, Forgejo, code-server, Excalidraw, Adminer, Matomo.
    • ImprovedSelecting every updatable server sends one request instead of one per server.
    • ImprovedUpdate buttons show a loading state and ignore double clicks.
  60. v0.2.7

    Ports of their own

    Each site gets its own container port independent of the host port, so several sites can share a server, and the `falak` CLI installs in one line.

    • NewContainer port decoupled from the host port: several sites can use the same port internally on one server.
    • NewOne-line install for the falak CLI on developer machines.
    • ImprovedA changed host port redeploys the site instead of leaving the old port live.
    • FixedDeleting a site stops its containers (and Compose project) instead of leaving them running.
  61. v0.2.6

    Domains that just work

    Every new service gets a working HTTPS address instantly, and custom domains tell you exactly which DNS records to add.

    • NewGenerated domains: <name>.<server-ip>.sslip.io with a Let’s Encrypt certificate and no DNS setup. nip.io or off per organization.
    • NewCustom domains show the A/AAAA records to add and check DNS live until the name points at your server. Cloudflare proxying is detected.
    • NewNetworking → Check DNS & TLS: records, a live check and the certificate actually served.
    • ImprovedCommands in flight survive agent restarts: desired-state commands are redelivered, one-shot steps fail with a clear message.
    • Fixedfalak-ctl detects changed config files through the container’s mount namespace and recreates exactly those services.
  62. v0.2.5

    Network logs and fleet upgrades

    Access logs per deployment, one-click agent upgrades across the fleet, and tighter release permissions.

    • NewNetwork Logs: every request your edge served, per release, with status and path filters.
    • NewUpgrade one agent or all of them from the panel, in batches, verified by checksum, with automatic stop on the first failure.
    • ImprovedLaravel web logs reach Loki with site attribution (multi-line stack traces merged).
    • ImprovedBuilders send run ids and heartbeats, so a restarted builder never leaves a deploy hanging.
    • FixedRelease directories are closed to other local users (0750 plus an ACL for the web server).
  63. v0.2.4

    A calmer, faster canvas

    Stacked floating panels with motion, deployment cards with logs, and groups, volumes and references on the canvas.

    • NewCanvas groups: frame related services, collapse them, move them as one.
    • NewVolume strips under services with persistent storage, and dashed edges for variable references and depends_on.
    • NewDeployment cards with the live release, failures and history; a stacked logs panel with build, deploy and network logs.
    • ImprovedProjects dashboard with starred projects, previews and health per environment.
  64. v0.2.3

    Hardened on real servers

    Fixes from deploying real apps on AWS: static sites, remote databases, multi-server health checks and more.

    • NewStatic sites get a 404 page or an SPA fallback instead of bare 404s, served from the release root.
    • NewRemote database access over the private network.
    • Improvedpnpm and yarn builds, build-time variables and custom build commands.
    • FixedMulti-server health checks and cloud public IP detection.
  65. v0.2.2

    Fast panel under load

    The panel now runs in FrankenPHP worker mode and agents long-poll their own service, so pages stay fast however many servers you add.

    • ImprovedPanel pages answer in 2.7–23 ms (median TTFB) with 50 agents connected (previously they queued for seconds behind long-polls).
    • NewA separate agent-api service for agents, installer and builders, with its own thread pool.
    • Newfalak-ctl status and doctor show PHP thread usage and flag saturated pools.
    • FixedGitHub App manifest conversion error handling.
  66. v0.2.1

    One-click GitHub

    Connect GitHub in one click with a private GitHub App: read-only access, no deploy keys, short-lived tokens per build.

    • NewGitHub App via the manifest flow, registered on your account or organization.
    • NewOne app-level webhook for push-to-deploy; installation tokens minted per build and never stored.
  67. v0.2.0

    Compose, templates and Octane

    Docker Compose deploys, a catalog of 18 one-click apps, Octane routing and smarter multi-server deploys.

    • NewDocker Compose sites from your repo or stored in Falak with history; images pinned by digest for exact rollbacks.
    • NewTemplates: n8n, Plausible, Umami, Ghost, WordPress, MinIO, Metabase and more, plus your own custom templates.
    • NewLaravel Octane (FrankenPHP, Swoole, RoadRunner) with stable per-site ports and zero failed requests on redeploy.
    • ImprovedDeploys wait for servers that are still provisioning instead of failing.
    • NewRuntime download mirrors for FrankenPHP, Node, Bun and Deno.
  68. v0.1.0

    First release

    The production installer, the Railway-style UI and everything underneath: servers, sites, deploys, databases and observability.

    • NewOne-command production install (Docker Compose, Let’s Encrypt) and falak-ctl for status, updates with rollback, backup and restore.
    • NewProject canvas with environments, services, a create picker, service panels and deploy views.
    • NewInfrastructure: server fleet, add-server flow, metrics, processes, firewall, network, terminal, SSH keys, recipes.
    • NewObservability: issues, traces, logs, heartbeats, alerts and a notification center.

Your servers are waiting.

One command installs Falak on a fresh Ubuntu or Debian host.

curl -fsSL https://falak.sh/install.sh | sudo bash -s -- --domain falak.example.com --email [email protected]