Open Source · Cloud Native · AI-native · Agent-native

Open-source control plane for CI/CD, AI and agent-native delivery.

Arcentra unifies CI/CD pipelines, AI training jobs, data workflows and agent runtime in one cloud-native command center. Self-host it, extend it, and run it across your infrastructure.

Get StartedView on GitHub
Apache-2.0Self-hostedKubernetes-nativeGitOps-ready
app.arcentra.io / overview
arcentra / overview
Control Plane
⌘ K
Pipelines
248
+12%
Deploy Frequency
6.4/d
+0.8
Lead Time
14m
-3m
MTTR
9m
-2m
Recent pipelines
last 24h
api · main2m ago
web · feat/auth-v28m ago
worker · main21m ago
infra · release/1.41h
Agent capacity
32 / 48
4 clusterslive
Quick start

Run Arcentra in under a minute.

$ docker
# Run Arcentra locally
docker run -p 8080:8080 \
  -v arcentra-data:/var/lib/arcentra \
  ghcr.io/arcentrix/arcentra:latest

# Open the control plane
open http://localhost:8080
Product preview

One command center for delivery operations.

Pipelines, environments, agents, runtime diagnostics and access control — every layer of your delivery stack lives in a single, self-hosted control plane.

See pipeline health, deployment frequency, lead time, MTTR and agent capacity.

app.arcentra.io / overview
arcentra / overview
Control Plane
⌘ K
Pipelines
248
+12%
Deploy Frequency
6.4/d
+0.8
Lead Time
14m
-3m
MTTR
9m
-2m
Recent pipelines
last 24h
api · main2m ago
web · feat/auth-v28m ago
worker · main21m ago
infra · release/1.41h
Agent capacity
32 / 48
4 clusterslive
Core workflow

Code → Build → Approve → Deploy → Observe → Recover.

One delivery loop, modeled as a connected graph instead of a stack of disconnected tools — with policy-gated approvals between every stage.

  1. Code
    step 1
  2. Build
    step 2
  3. Approve
    step 3
  4. Deploy
    step 4
  5. Observe
    step 5
  6. Recover
    step 6
build

Build

Pipelines, runners, artifacts and execution history.

approve

Approve

Required approvals, change windows and policy gates between stages.

deploy

Deploy

Environments, releases, GitOps and rollback controls.

agents

Agents

Distributed execution across clusters, GPU pools and private networks.

observe

Observe

Runtime health, pipeline status, logs and diagnostics.

secure

Secure

Identity, access, secrets and policies.

Workloads

One control plane for every workload.

CI/CD pipelines, AI training jobs and data workflows share the same scheduler, agents, secrets and policies. No more duct-taping three different platforms together.

CI / CD

Build, test and ship code

Run pipelines, build artifacts, gate releases and deploy services across your environments.

  • $build · api · main
  • $test · web · pr-2148
  • $deploy · staging → production
AI / ML

Train, evaluate and serve models

Schedule training and fine-tuning jobs on GPU pools, run evals, manage prompts and ship inference endpoints.

  • $train · llm-7b · finetune-v3
  • $eval · ragas-suite · nightly
  • $serve · inference · canary
Data

Run ETL and big-data jobs

Orchestrate Spark, Ray and Airflow workloads. Schedule batch, streaming and scheduled jobs alongside CI/CD.

  • $etl · daily · 03:00 UTC
  • $spark · feature-store · backfill
  • $stream · kafka → warehouse
Why Arcentra

Modern delivery stacks are fragmented.

CI lives in one place. Deployments live somewhere else. Agents are hard to operate. Secrets are scattered. Observability is disconnected. Arcentra connects them into one open-source control plane.

Before Arcentra
  • Scattered pipelines across vendors
  • Manual deployment operations
  • Hard-to-manage runners
  • Hidden failure context
  • Separated access control
With Arcentra
  • Unified pipeline view across teams
  • Controlled release workflow with policies
  • Agent runtime, observable in one place
  • Runtime diagnostics next to the run
  • Workspace-level identity and secrets
Architecture

Transparent, layered, cloud-native.

A control plane you can read, audit and extend. Three layers, each replaceable, each running on infrastructure you own.

flow · data planemTLS · gRPC · webhooks
Git ProviderGitHub / GitLabArcentra Control PlaneAPI · Scheduler · Workspace · Policy · GitOpsAgentsrunners · workersK8s / VMexecution targetsDatabase / Vaultstate · secretsObservabilitylogs · metrics · traces
Layer

Control Plane

Workspaces, RBAC, scheduling, GitOps reconciliation, approval gates and policy evaluation.

APISchedulerWorkspacePolicyGitOpsApprovalsAudit
Layer

Execution Layer

Distributed runners and GPU pools that execute CI, AI and data jobs on your clusters, clouds and private networks.

AgentsRunnersGPU poolsWorkersTask runtimeSandbox
Layer

Integration Layer

Pluggable adapters that connect Arcentra to source control, schedulers, ML tooling and your existing infrastructure.

GitHubGitLabKubernetesSpark / RayAirflowMLflowSecretsObservability
Use cases

Built for the teams shipping the platform.

use case

For platform teams

Build an internal delivery platform without locking into a single CI vendor.

use case

For DevOps teams

Unify pipelines, deployments, agents and runtime diagnostics.

use case

For AI / data teams

Schedule training, evals, ETL and streaming jobs on shared GPU and CPU pools.

use case

For cloud-native teams

Run delivery workflows across Kubernetes clusters and private networks.

use case

For open-source teams

Self-host your CI/CD control plane with transparent architecture.

use case

For enterprise teams

SSO, audit logs, approval policies and air-gapped deployments out of the box.

Open source

Open source by design.

Arcentra is built in the open for teams who want control, extensibility and self-hosted delivery infrastructure. Read the code, file an issue, ship a patch — or fork it.

Open-source firstSelf-hostedExtensibleCloud-native
GitHubContributeRoadmap
~/arcentrix/arcentra
$ git clone git@github.com:arcentrix/arcentra.git
$ cd arcentra
$ make dev
→ control plane on :8080
→ agent runtime on :9090
→ docs on :3001
# everything you need to hack on Arcentra.

Ready to build your delivery control plane?

Start locally, connect your agents, and ship your first pipeline.

Get StartedView on GitHub