Levelrail is approaching its stable release: a single node is stable, and every feature is labelled stable, beta or flagged by the evidence behind it. See what is verified.What is verified Questions? Join the Discord
Deliberately minimal: no control plane, database or agent, running over SSH from your machine or CI.
As of 2026-11-05. Statements about Kamal come from its own public documentation, linked on each row. Spot something outdated? Tell us and we will fix it.
38capabilities compared, from each project's own docs
1rows where Kamal is ahead
8rows where Levelrail is ahead
The short verdict
Kamal is good at
Kamal is good at: as few moving parts as possible; and pairs with kamal-proxy for zero-downtime cutover; and no dashboard needed, which suits some teams; and MIT licensed with no paid tier, a built-in deploy lock, hooks that can abort a deploy, and native OpenTelemetry for deploy logs.
When to choose Kamal
Choose Kamal when you want the fewest possible moving parts and do not need a dashboard.
Where Levelrail differs
Levelrail differs here: desired state lives in a database, with a reconciler recording a status reason after every pass; and agents dial out, so CI uses a scoped API token instead of holding SSH keys; and adds a dashboard, deploy history, rollback and built-in observability, and is heavier by design. Levelrail has a shorter track record: a single node is the well-tested path and multi-node is labelled beta, so its track record is much shorter than Kamal's.
Where Kamal is ahead
1
Multi-node
YesKamal: SSH to each server.
PartlyLevelrail: partly, join, cordon and drain with real container relocation verified across two local Docker daemons; WireGuard mesh and cross-host transport still unverified. Resource-aware placement is beta.
Where Levelrail is ahead
8
No SSH: agents dial out
YesLevelrail: Agents dial out over mTLS; no inbound ports.
NoKamal: no, SSH from your machine or CI to each server.
Level-triggered reconciler with status reasons
YesLevelrail: Desired state converged, status condition with reason per pass.
NoKamal: no, none: it runs Docker commands per deploy.
Metrics and log search built in
YesLevelrail: Node-local metrics at 15s and full-text log search.
NoKamal: no, bring your own.
One-click rollback with pinned images
YesLevelrail: Prior images pinned so cleanup cannot remove them. A 5xx spike or crashloop after cutover rolls back automatically (live-tested); the latency signal is beta.
PartlyKamal: partly, rollback to chosen version; old containers pruned after 3 days.
Encrypted secrets for env vars
YesLevelrail: Envelope-encrypted, masked in UI, injected at container create.
PartlyKamal: partly, secrets from files or password managers; host env file, no encryption documented.
IAM, audit log and 2FA in the free edition
YesLevelrail: Allow/Deny IAM policies (stable, live e2e), passkeys and TOTP, and an audit log with a verifiable hash chain, CSV export and forwarding, all in the one edition.
PartlyKamal: partly, per-server audit log (.kamal/app-audit.log) of deploy steps, last 50 lines via kamal audit; no users, roles or 2FA.
Audit log of every action
YesLevelrail: Every mutating request recorded; verifiable hash chain, CSV or NDJSON export, retention, streaming to a webhook or S3 bucket.
PartlyKamal: partly, per-server .kamal/app-audit.log of deploy steps, read with kamal audit; entries carry a user tag but the page does not say which person; hooks expose KAMAL_PERFORMER for custom reports.
Deploy freezes and policies
YesLevelrail: Cron-defined freeze windows per app with an audited override reason; also gate image auto-update and scheduled deploys.
PartlyKamal: partly, kamal lock acquire can block deploys during a maintenance window and a non-zero exit from pre-deploy or other hooks aborts the command (hooks can be skipped with --skip_hooks); no scheduled freeze windows or policy engine.
38 capabilities side by side
Yes, partly and no are what each project's own documentation states. "Not stated" means their docs do not say, which is not the same as no.
Levelrail maturity tags (beta, stable) come from its feature status page.
Partlykamal lock acquire can block deploys during a maintenance window and a non-zero exit from pre-deploy or other hooks aborts the command (hooks can be skipped with --skip_hooks); no scheduled freeze windows or policy engineSource for Kamal, Deploy freezes and policies
PartlyApps with more than one replica are round-robin balanced by default, with failed-replica skip and retry. The configurable balancer (health checks, sticky sessions, weights, drain) is experimental, behind a flag, verified on one machine only. HTTP onlySource for Levelrail, Load balancing across replicas
PartlySelf-signed by default; public ACME (HTTP-01) has one recorded run on a live VPS. Renewal near expiry and DNS-01 wildcards are not exercised live; CAs needing EAB are unsupportedSource for Levelrail, Automatic TLS certificates
PartlybetaJoin, cordon and drain with real container relocation verified across two local Docker daemons; WireGuard mesh and cross-host transport still unverified. Resource-aware placement is betaSource for Levelrail, Multi-node
YesbetaMCP server (beta) with over 240 tools in toolsets an agent enables on demand (256 at last count, 21 in the default session); in-app chat is behind a flagSource for Levelrail, MCP / AI assistant integration
Partlyconfig/deploy.yml declares app, roles, proxy, accessories and servers and kamal setup/deploy applies it; commands run per deploy, no state reconciliation or drift detection documentedSource for Kamal, Declarative apply of the whole platform
PartlybetaOIDC sign-in (17 named providers plus generic OpenID Connect) is on by default once configured. SAML and SCIM are beta, behind a toggle and off by default; SAML ran against Keycloak and SimpleSAMLphp in a local lab, not Okta or EntraSource for Levelrail, SAML or SCIM single sign-on
PartlyPer-server .kamal/app-audit.log of deploy steps, read with kamal audit; entries carry a user tag but the page does not say which person; hooks expose KAMAL_PERFORMER for custom reportsSource for Kamal, Audit log of every action
Two-person deploy approvals
YesstableA different person with the deploy ability must approve; the requester cannot approve their own. Approval also works from Slack or DiscordSource for Levelrail, Two-person deploy approvals
Not stated in their docs
Kamal is ahead on this row
Levelrail is ahead on this row
All ten projects are on the full matrix. Product names and logos belong to their owners and are used for identification only.
Migrating from Kamal
No importer for Kamal. Point Levelrail at the same images and recreate configuration.
Deliberately minimal: no control plane, database or agent, running over SSH from your machine or CI. How Levelrail differs: Desired state lives in a database, with a reconciler recording a status reason after every pass. Agents dial out, so CI uses a scoped API token instead of holding SSH keys. Adds a dashboard, deploy history, rollback and built-in observability, and is heavier by design. Choose Kamal instead when you want the fewest possible moving parts and do not need a dashboard. Levelrail has a shorter track record, so check the comparison for what it does not do yet.
When should I choose Kamal instead of Levelrail?
Choose Kamal when you want the fewest possible moving parts and do not need a dashboard.
Can I move from Kamal to Levelrail?
No importer for Kamal. Point Levelrail at the same images and recreate configuration.
Is Levelrail ready to replace Kamal in production?
Formats can change: APIs, the app spec and the on-disk layout can change between releases. A single node is stable; multi-node is in beta and the WireGuard mesh is not battle-tested. Check the feature status page before moving anything you cannot afford to lose, and keep Kamal running alongside until you trust the result.