Kibi

A lightweight, multi-cluster Elasticsearch workbench.

Kibi is for teams that use Elasticsearch as part of their product and want a fast operational interface without deploying a full Kibana instance beside every cluster. Run one Kibi server, register the clusters it can reach, and move between them from a single focused UI.

Kibi deliberately keeps authentication native: each person connects to a cluster with that cluster's Elasticsearch username/password or API key. Elasticsearch remains the authority for permissions, and Kibi keeps the credential only in its server-side session.

What you can do

• Work across many Elasticsearch 8+ clusters from one Kibi deployment.
• Sign in per cluster using native Basic authentication or an Elasticsearch API key.
• See cluster health, compatibility details, node roles and resource pressure.
• Inspect indices, mappings, settings, aliases, document counts, and storage.
• Follow primary and replica shards in a node placement map, including recovery and unassigned state.
• Run Query DSL searches with readable expandable documents or the raw Elasticsearch response.
• Use a REST console with searchable operational recipes, browser-local history and favourites, and production confirmation for mutating calls.
• Create, replace, partially update, or delete individual documents from an index workspace, again with a production guardrail.

Kibi is intentionally not an identity provider and does not replace Elasticsearch security. It is a focused back-office interface for the APIs your search-engineering team already uses.

Screenshots

Cluster overview

One workbench surfaces health, index and shard counts, connected user, Elasticsearch compatibility details, and quick paths into operational views.

Index workspace

Open an index to inspect its mapping, settings, aliases, statistics, and documents with a scoped Query DSL browser.

Search response, your way

Search results can be read as expandable documents or switched to the complete Elasticsearch JSON response when debugging needs the exact payload.

REST console

Run a JSON Elasticsearch request against the selected cluster with native credentials and production safeguards.

Use it locally

Prerequisites: Docker, Go 1.22+, and Node.js 20+.

make up       # starts local development and production Elasticsearch 8 clusters
make api      # starts Kibi's Go API on http://localhost:8080
make web      # starts the Vite UI on http://localhost:5173

Open http://localhost:5173, select Development (local), and sign in with elastic / changeme-dev. The development environment is seeded with a small catalog, customer, order, and support-ticket dataset so every view has meaningful data. The bundled production cluster is kept empty and uses elastic / changeme-prod.

To restore the local sample data after the cluster is already running:

make seed

To stop the local Elasticsearch services:

make down

make build compiles both applications. On some macOS Go installations, the external linker used by the Makefile avoids a system-loader issue.

Connect remote clusters

Kibi's API server must have network access to every registered Elasticsearch endpoint, normally through a private network or VPN. Give the server its catalogue through KIBI_CLUSTERS:

export KIBI_CLUSTERS='[
  {
    "id": "search-production",
    "name": "Search production",
    "endpoint": "https://es-prod.example.internal:9200",
    "environment": "production"
  },
  {
    "id": "search-staging",
    "name": "Search staging",
    "endpoint": "https://es-staging.example.internal:9200",
    "environment": "development"
  }
]'
make api

Use config/clusters.example.json as a complete starting shape. Users choose a cluster and authenticate directly against it; their access in Kibi is therefore constrained by the Elasticsearch roles, users, and API keys you already manage.

For a shared deployment, host the API and compiled frontend behind your normal internal ingress. Set KIBI_COOKIE_SECURE=true when serving HTTPS. Kibi's session storage is currently memory-only, so credentials disappear when the API process restarts and deployments should use a single API instance until shared session storage is introduced.

Technology

• Go standard library HTTP API and Elasticsearch REST proxy
• React + TypeScript + Vite frontend
• Elasticsearch 8.x for the local demo environment
• Docker Compose for local infrastructure

Repository layout

backend/     Go API and its tests
frontend/    React workbench
config/      Deployable cluster-catalogue examples
dev/         Local-only Elasticsearch demo tooling
docs/        Product direction and screenshots

Development notes and longer-term direction live in docs/product-direction.md.

Website

The public product website lives in website. It is a zero-build static site ready for Cloudflare Pages; see website/README.md for the Git-connected deployment settings.