Licensed to be used in conjunction with basebox, only.
// reference
Version matrix
Applies to
Product: Server (Cloud for reference) · Audience: Platform Operator · All
Which Helm chart version ships which basebox version, which components are bundled and with which infrastructure versions basebox was verified. In basebox Cloud, basebox keeps the environment up to date; there the matrix is relevant for reference only.
Current release
| Version | |
|---|---|
Helm chart basebox.ai |
0.3.32 |
| basebox (app version) | 1.8.8 |
| Chart source | oci://gitea.basebox.health/basebox-distribution/helm/basebox.ai |
The chart version is the value you pass to helm upgrade --install … --version; the app version is what the interface shows under "About". Installation guide: Using Helm charts.
What the umbrella chart bundles
| Component | Role | Image / source |
|---|---|---|
| frontend | Web interface (Vue 3) | gitea.basebox.health/basebox-distribution/frontend |
| AISRV | Application server, GraphQL/REST/OpenAI-compatible API, MCP gateway | …/aisrv |
| storesrv | Store service (GraphQL) | …/storesrv |
| ragsrv | RAG vector service (PostgreSQL + pgvector) | …/ragsrv (gpu-latest / cpu-latest) |
| ragsrv-support | Model service for extraction, OCR, embeddings, STT | …/ragsrv-support |
| idp | Keycloak identity provider | …/idp |
| inference | Bundled LLM inference server (vLLM) | …/inference |
| CloudNativePG operator | PostgreSQL cluster per service | ghcr.io/cloudnative-pg/postgresql:16-standard-bookworm (database image) |
| MCP connectors | Optional, enabled per connector | see MCP connectors with Helm |
Read the exact image tags of an installed release from the cluster:
kubectl -n basebox get deployments -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{.spec.template.spec.containers[0].image}{"\n"}{end}'
Release history
| basebox | Date | Highlights | Chart version |
|---|---|---|---|
| 1.8.x | 2026-08-19 | Connectors (MCP), web search, thinking modes, sources and citations, audit viewer, organisation-wide instructions, new design | 0.3.32 (current) |
| 1.7.0 | 2026-04-25 | OpenAI-compatible API, group management, Swagger UI, message of the day, enhanced audit logging, Keycloak 26.5.6 | – |
| 1.6.x | 2026-03-07 | App sharing, reasoning models, speech-to-text, vLLM as inference engine (TGI in maintenance mode) | – |
| 1.5.x | 2025-12-12 | Licence management for Server, split-screen UI, documentation site online | – |
| 1.1.x | 2025-10-28 | First 1.x release: new frontend (Vue.js), new extraction pipeline, OCR | – |
Details per release: Changelog.
Migration note from 1.8.6
basebox 1.8.6 adds the AISRV database migration V32. Returning to an AISRV version older than 1.8.6 requires restoring the AISRV database backup taken before the upgrade together with the earlier Helm release. Back up before every upgrade – see Updates.
Verified infrastructure versions
These versions were verified by basebox with the Server Preparation Guide (as of February 2026). They are test benches, not minimum requirements; the chart's minimum requirements are listed below.
| Component | Verified version |
|---|---|
| Ubuntu | 24.04 LTS |
| NVIDIA driver | 580.126.09 |
| CUDA Toolkit | 13.0 (12.9 as fallback) |
| Docker | 29.2.1 |
| Kubernetes | 1.33.7 |
| Helm | 3.20.0 |
| Calico CNI | 3.31.3 |
| NVIDIA GPU Operator | latest via Helm |
Chart minimum requirements: Kubernetes 1.23+, Helm 3.x, dynamic volume provisioning, an ingress controller, NVIDIA GPU Operator (or equivalent) on GPU nodes – see Helm chart overview.
Verified inference combinations
| basebox | Inference runtime | Model | Context |
|---|---|---|---|
| 1.7.1 | vLLM 0.15.0 (external endpoint) | openai/gpt-oss-20b |
32,768 |
Further verified models and configurations: Tested models and Reference configurations. GPT-OSS models require vLLM 0.10.1 or newer.
Compatibility rules
- Chart and app version belong together. Always install the chart version that belongs to the desired basebox version; do not mix individual charts of different vintages.
- Upgrades in order. Databases first (for schema changes), then backend services, IDP, inference, frontend last – this is how
helm upgradewith the umbrella chart proceeds. - Downgrades need a backup. Database migrations are forward-only.
- Inference is decoupled. The inference runtime can be updated independently of basebox as long as it continues to serve the OpenAI-compatible API (
/v1/models,/v1/chat/completions) and reports the same model identifier.
Need help? Contact support