langship.sh

v0.1.0 · Apache 2.0 · Self-hosted

Maintained by the Lyzr team ↗

Vercel for AI agents.

Build and ship
any agent.

Any framework. Any runtime. Build with LangChain, LlamaIndex, CrewAI, AutoGen, or your own SDK — then deploy, govern, and operate them on Kubernetes, Bedrock AgentCore, or Vertex AI Agent Engine. Self-hosted, end to end.

Deploy a sample agentRead the architecture
~/install.sh
# 1. install
curl -fsSL https://get.langship.sh | sh

# 2. point at your project
langship link my-agent

# 3. ship
langship deploy --env=prod
3
runtimes
8
node types
GitOps
native
BUILDSCANDEPLOYOBSERVEN. 01 — DEPLOY

fig. 01 — control plane

A walkthrough

  1. 01GitHub push lands on main.
  2. 02Build, scan, SAST run in parallel.
  3. 03Approval gate held for human review.
  4. 04Deploy to Vertex Agent Engine.
  5. 05SLO watcher armed for 5 minutes.

Release flow

devdeployed
v2.4.3a71a73d
k8s · local
deployed 4m ago · evals=passed
auto-promote
stagingpending
v2.4.3a71a73d
bedrock-agentcore
1 / 2 approvals · waiting on @lyzr-leads
approval pending
proddeployed
v2.4.1608472f
vertex-agent-engine
deployed 2d ago · slo=healthy

Environments

Same agent.
Different gates per stop.

Each environment owns its own pipeline, runtime, approval policy, and eval gate. Promotion between them is explicit, gated, and follows your branching strategy.

Branching

trunk-based·env-branches·release branches·custom
dev
fast feedback
01
staging
soak tests
02
prod
real traffic
03
Runtimek8s · local
bedrock-agentcore
vertex-agent-engine
Branchmain
main
release/*
Auto-promoteyes
no
no
Approval—
1 of eng-leads
2 of platform-team
Eval gatesmoke (60s)
regression (5m)
full + cost (15m)
Budget cap$5 / day
$50 / day
$2k / day
Audit retention30d
90d
365d
dev
auto
staging
approval
prod
+ rollback follows the same path

CI/CD as a graph

Drag, drop, ship.
YAML is the source of truth.

Visual canvas in the UI, declarative file in git. PR review, audit trail, rollback via git revert.

Triggergithub
Buildbuildkit
Scantrivy
SASTsemgrep
Approvalhuman
Deployvertex
Promote→ prod
.langship/pipelines/prod.yamlread-only
# committed to git, edited via UI or your editor
name: ship-prod
trigger: { kind: github, branch: main }
nodes:
  - { type: build,    engine: buildkit }
  - { type: scan,     engine: trivy, fail-on: [HIGH, CRITICAL] }
  - { type: sast,     engine: semgrep }
  - { type: approval, approvers: [eng-leads] }
  - { type: deploy,   runtime: vertex-agent-engine }
  - { type: promote,  to: [staging, prod] }

Runtime-agnostic

One agent. Any runtime.

01

Kubernetes

Self-hosted, any cloud. Helm chart for prod, Compose for local dev.

OSS · Self-hosted
02

AWS Bedrock AgentCore

Managed agent runtime on AWS. Same pipeline, same governance.

Managed · AWS
03

GCP Vertex AI Agent Engine

Managed agent runtime on Google Cloud. Same governance applies.

Managed · GCP

Works with

LangChain·LangGraph·LlamaIndex·CrewAI·AutoGen·Pydantic AI·DSPy·Haystack·Raw SDK

Get the platform

Bring your agent.
We'll handle the boring parts.

Talk to a maintainer
  • Helm chartprod
  • Docker Composelocal dev
  • CLImacOS · linux · windows
  • LicenseApache 2.0
  • Supportgithub · discord · email