Skip to content

Ship Go and Rust apps.Watch them. Roll them back.

One CLI that builds, deploys, and supervises them on infrastructure you control, with health-gated traffic switching for proxy-based deploys. A built-in MCP server lets coding agents like Claude Code and Codex drive that same pipeline, plan-bound and approval-gated.

Read the docs
runtime
Go · Rust
deploys
blue-green
rollback
one command
phelix · example sessionstopped
phelix statusproxy → ·
apiGov18:8080stopped

no processes

Released Phelix hosts: Linux and macOS. Builds Go and Rust with your host toolchain.

Source in. Versioned binary out.

A successful Go or Rust build creates a numbered candidate with its encrypted environment snapshot. It becomes current only after the selected deployment succeeds.

Example outputgo 1.27 · ~/src/api
·Detect project
·Compile native binary
·Create numbered candidate and encrypted snapshot
·Candidate ready

$phelix build

Every version records a build report: compiler, duration, cache status, size, and checksum, compared against the five previous comparable builds. Matrix builds cross-compile toolchain × platform combinations in one command.

Five ways to ship a version.

Classic, blue-green, rolling, canary, and progressive share one versioned lifecycle. Proxy-based strategies move traffic only after the new version passes its health gate; classic is stop → start.

deployment exampleatomic target switch
$ phelix rebuild api --blue-green
  1. start v19 on the inactive slot, pending
  2. v19 health gate, pending
  3. switch proxy target, pending
  4. commit serving state · promote v19, pending
  5. drain v18, pending

Builds a new binary, starts it on the inactive slot, waits until healthy, then atomically switches the proxy target.

If the candidate fails its health gate, the deploy aborts and the active version keeps serving.

Push to deploy the exact commit.

Phelix accepts HMAC-authenticated Git push webhooks, deploys the exact commit that was pushed, and hands off to the same rebuild pipeline. Your configured strategy, health gates, versioning, and rollback history all still apply.

webhook example127.0.0.1:9746
The pushed commit SHA is always the deploy target, never the branch tip. Every accepted delivery is recorded in a durable ledger before it is acknowledged, so a redelivery never triggers a second rebuild.

MCP

Let agents deploy. Keep the final say.

Phelix ships a built-in MCP server, so coding agents (Claude Code, Codex, any MCP client) drive the same build, deploy, and rollback pipeline as the CLI over local stdio. But an agent can only create a plan and request its apply: execution stays denied until a human approves that exact plan by its content hash. The agent proposes; you authorize; Phelix executes.

Read the MCP guide
mcp examplephelix mcp serve --stdio
The MCP adapter is a transport only: no new engine, no embedded AI, no network listener. On an unconfigured host, apply is denied outright; opting in still binds every approval to one plan's id and content hash, so an approval can never authorize a different plan.

One lifecycle. Less glue.

Release state is often split across build scripts, process managers, health checks, proxy rules, and rollback notes. Phelix keeps one version identity and promotion rule across that lifecycle.

without phelix

with phelix

phelixone operational layer
buildrundeployhealthlogsrollbackmonitor

application → candidate → health gate → current → rollback

The layer between your application and your servers.

Phelix starts where source reaches a managed host. It owns the build, the process, health-gated traffic, and the retained way back. Git, CI, and the server itself stay yours.

  1. your application

    Go · Rust source

  2. phelix
    buildrundeployhealthlogsrollbackmonitor

    cli + per-host agent

  3. your servers

    Linux · macOS · processes · systemd · proxy · network

  4. phelix dashboardopt-in

    status · metrics · logs · events

Phelix sits between your application and your servers: it builds, deploys, and health-checks the application, then supervises processes and routes traffic on the servers. Connected servers report telemetry over authenticated TLS to the optional Phelix dashboard.

outside phelix

Git and CI prepare the source. You provision the server, network, and TLS.

request path

User traffic reaches a host-local HTTP proxy, which routes it to the healthy serving process.

client → :8080 proxy → healthy instance

optional connection

Login is optional and everything works locally without it. Connected agents report status, metrics, and logs; remote actions target one selected agent.

Phelix operates applications on infrastructure you already control. It does not provision the server.

The lifecycle, end to end.

Build, process control, logs, health, optional connected observability, resource status, and an agent-facing MCP server share the same versioned application context.

build

One-command builds

Compile a Go or Rust app into a numbered native candidate.

run

Run and restart

Start, stop, and supervise processes through explicit application state.

stop

Graceful stop

Stop a running application by ID and allow configured shutdown handling.

logs

Continuous local logs

Tail local process logs until interrupted.

remote

Connected observability

Inspect telemetry reported by one authenticated agent when the optional connection is configured.

metrics

Resource status

Inspect PID, uptime, memory, and CPU per application.

mcp

Agent-ready MCP server

Expose the same build and deploy pipeline to coding agents over local stdio, plan-bound and approval-gated, with no new execution path.

  • Any Go version
  • Any Rust version
  • Linux release
  • macOS release
  • systemd on Linux
  • Go and Rust images
  • Compose generation
  • connected agent
  • MCP server

Watch it run.

Tail logs locally, or inspect example logs and resource status reported through a connected agent. This panel is simulated.

phelix log api --tailSimulated example
19:42:01GET/api/users20014ms
19:42:02GET/api/products20021ms
19:42:03POST/api/orders20148ms
19:42:04GET/api/users/4220016ms
CPU
34%
MEMORY
182 MB
UPTIME
7d 14h
STATUS
healthy

Simulated metrics: CPU 34 percent, memory 182 MB, status healthy.

Every server, one surface.

Connected agents can report CPU, memory, uptime, process state, health snapshots, and logs over authenticated TLS. Select an example server to inspect it.

Example connected fleettelemetry · every 2s

api @ server-01healthy
version
v18
port
8080
cpu
34%
memory
182 MB
uptime
7d 14h
region
fra1
agent
connected
restarts
0

Fleet visibility spans connected agents; remote actions target one selected agent.

Failure stays explicit.

Repeated health-check failures can trigger bounded restart attempts. During deployment, an unhealthy candidate stays off traffic; optional automatic rollback can restore the last known-good promoted version.

recovery mechanismsSimulated example

Health-triggered restart

Repeated failures schedule bounded restart attempts with an attempt cap.

attempt cap reached → restart skipped · operator action required

Deployment recovery

An unhealthy candidate stays off traffic; restoration can return traffic to the last known-good promoted version.

candidate unhealthy → kept off traffic → optional restore last-known-good → traffic restored | restoration failed → degraded · manual intervention required

Rollback is a first-class command.

Choose a retained native version. Phelix resolves it, starts it through the application's current deployment topology, runs topology-specific checks, and records the outcome.

phelix versions api
$ phelix rollback api --to v17example topology · proxy-based

This example shows a proxy-based rollback switching traffic after health passes. Phelix reuses the application's configured topology; classic rollback uses stop → start and can optionally run post-rollback verification.

Built and run by

Anophel logoAnophel
“We build Phelix and run it in production at Anophel: it deploys, supervises, and rolls back our own Go and Rust services across several servers from one CLI.”
Mohammad AbdorrahmaniMohammad AbdorrahmaniFounder, Anophel
“The real-time monitoring and centralized logging features have made debugging and maintenance so much easier across our infrastructure.”
Mohammad MasoudiMohammad MasoudiCEO · Full-Stack Developer, Becca

Pay for capacity, not seats.

Flat monthly rates per plan. Settle in USD or in crypto at the current rate. Every paid plan includes zero-downtime deploys and multi-server monitoring.

billing · plansusd · btc · eth · usdt · doge · ton

Questions engineers actually ask.

Short answers here, full detail in the docs.

One binary between git push and production.

Install the CLI, verify the host, and follow the quick start to build your first application.

Read the docs