Helm charts maintained by upstreamyard, for our own container images and for projects' official images.
| Chart | App | Image |
|---|---|---|
activemq |
Apache ActiveMQ Classic message broker (JMS, AMQP, STOMP, MQTT): single broker or high-availability master/slave, metrics | upstreamyard/activemq |
aegra |
Aegra, a self-hosted alternative to LangGraph Platform | upstreamyard/aegra |
automq |
AutoMQ, Kafka-compatible streaming with messages stored in S3 (diskless Kafka) | automqinc/automq (official) |
cube |
Cube, semantic layer and data API for BI and AI, with Cube Store | cubejs/cube, cubejs/cubestore (official Cube images) |
docling-serve |
Docling Serve, the Docling document conversion API (PDF, DOCX, PPTX to Markdown/JSON for RAG), with RQ workers and KEDA autoscaling | official ghcr.io/docling-project/docling-serve-cpu and CUDA images |
discourse |
Discourse, the open-source community forum | upstreamyard/discourse |
dolt |
Dolt, the version-controlled MySQL-compatible SQL database, with primary/standby high availability | dolthub/dolt-sql-server (official) |
electric |
Electric, a sync engine that streams subsets of your PostgreSQL data to web and mobile apps in real time | electricsql/electric (official) |
falkordb |
FalkorDB graph database for knowledge graphs and GraphRAG, single instance or HA with Sentinel failover | official falkordb/falkordb-server |
kvrocks |
Apache Kvrocks, Redis-compatible key-value database on RocksDB, with Sentinel high availability | apache/kvrocks (official) |
localai |
LocalAI, the self-hosted OpenAI-compatible API for LLMs, embeddings, images and speech on CPU or GPU: single server or distributed frontends and GPU workers, API key and admin account by default | official localai/localai (CPU and GPU variants) |
logto |
Logto, identity and access management (OIDC, SSO) for apps and SaaS | svhd/logto (official Logto image) |
conductor |
Conductor OSS, a workflow orchestration engine originally built at Netflix | upstreamyard/conductor |
nemo-guardrails |
NVIDIA NeMo Guardrails, an OpenAI-compatible LLM guardrails server (jailbreak, content safety, PII and dialog rails) | upstreamyard/nemo-guardrails |
parse-server |
Parse Server, the open-source backend for mobile, web and IoT apps | parseplatform/parse-server (official) |
saleor |
Saleor, the headless GraphQL commerce platform, with its dashboard | official ghcr.io/saleor/saleor and ghcr.io/saleor/saleor-dashboard |
plano |
Plano AI gateway and LLM proxy: one OpenAI-compatible API for many providers, starts without internet access, non-root, API keys kept from the Envoy admin API, PostgreSQL state and Redis cache for several replicas, zero-downtime upgrades | official katanemo/plano |
postal |
Postal, a mail delivery platform for sending and receiving email | official ghcr.io/postalserver/postal |
shardingsphere-proxy |
Apache ShardingSphere-Proxy, a database proxy for sharding, read/write splitting, encryption and masking in front of MySQL and PostgreSQL | upstreamyard/shardingsphere-proxy |
supertokens |
SuperTokens, the open-source authentication server (sessions, passwordless, social login, MFA) | supertokens/supertokens-postgresql (official) |
tigerbeetle |
TigerBeetle financial transactions database: fault-tolerant 3 or 6 replica cluster, safe replica recovery, rolling upgrades, Prometheus metrics | official ghcr.io/tigerbeetle/tigerbeetle |
xxl-job |
XXL-JOB admin, the distributed job scheduler for Java services | xuxueli/xxl-job-admin (official XXL-JOB image) |
helm repo add upstreamyard https://upstreamyard.github.io/helm-charts
helm repo update
helm install aegra upstreamyard/aegra --set database.url='postgresql://user:password@host:5432/aegra'Charts are also published as signed OCI artifacts:
helm install aegra oci://ghcr.io/upstreamyard/charts/aegra --set database.url='...'
cosign verify ghcr.io/upstreamyard/charts/aegra:<chart-version> \
--certificate-identity-regexp '^https://github.com/upstreamyard/helm-charts/' \
--certificate-oidc-issuer https://token.actions.githubusercontent.com- Daily (
bump.yml): if a packaged project (listed per chart in.github/bump/) released a new version and the image for it is published, a bot opens a PR that updatesappVersionand bumps the chart's patch version. - On every PR (
lint-test.yml): lint, README check, then a real install in a kind cluster against PostgreSQL (pgvector), Redis and the services the changed chart needs, withhelm test. - Bot PRs merge themselves once
lint-testpasses. If it fails, the PR stays open for a human. PRs that also change files bundled with a chart (e.g. a database schema, viascripts/bump-hooks/<chart>.sh) always wait for a review. - On merge to
main(release.yml): a GitHub Release, theindex.yamlon GitHub Pages, the OCI chart on GHCR (cosign-signed) and Artifact Hub metadata. Artifact Hub re-scans the repository on its own.
Open a PR against main. Bump the chart version for every chart change, and regenerate the chart README with helm-docs: helm-docs --chart-search-root charts.
