9.4.0

Summary

Feature release adding RS_SETUP_PROFILE tiers for lean deployments against external Elasticsearch/OpenSearch clusters (as few as 4 primary shards). Also includes Render one-click deploy with minimal-profile defaults, nil-safe guards when optional meta indices are skipped, and a binary publish workflow fix for Packer AMI builds. Bumps the default version to 9.4.0.

---

New features

Setup profiles (RS_SETUP_PROFILE)

Control which meta (system) Elasticsearch indices are created at startup. Unset defaults to full (backward compatible).

Profile	Primary shards (fresh install)	Indexes created
minimal	4	.users, .permissions, .pipelines, .pipeline_vars
standard	12	minimal + .synonyms, .searchrelevancy, .logs, .pipeline_logs, .pipeline_invocations, .analytics, .user_sessions, .analytics_preferences
full	~35–48	All meta indexes (current behavior)

• Central registry — util/setup_profile.go with ShouldCreateMetaIndex() gates index creation across all plugins.
• Shard reduction — MetaIndexShards() forces 1 primary shard per index on minimal and standard; full keeps per-plugin defaults.
• JWT on lean profiles — .publickey is not created on minimal / standard; set JWT_RSA_PUBLIC_KEY_LOC for JWT auth instead.
• Example config — config/minimal.env.example for local testing.
• Smoke tests — scripts/smoke-minimal.sh validates auth, pipelines, pipeline vars, and ES proxy under the minimal profile.

Render one-click deploy

• render.yaml blueprint — Docker web service with RS_SETUP_PROFILE=minimal, health check at /arc/health.
• README — Deploy to Render button and pairing guide with Aiven free OpenSearch for a free end-to-end stack.
• ES_CLUSTER_URL, USERNAME, and PASSWORD are prompted at deploy time (sync: false).

---

Fixes

• Nil-safe lean profiles — Guard against panics when optional meta indices are not created:
  • Nodes plugin: skip .nodes index init and cron jobs; /arc/_health returns health: ok without node counts.
  • Pipelines: skip invocation record writes when .pipeline_invocations is absent.
  • Analytics / applycache middleware: skip persistence and cache-stats rollover when respective indices are absent.
• Binary publish workflow — Restore rs-linux-ami.zip / rs-linux-cluster.zip asset names expected by Packer AMI upgrade.sh.

---

Configuration

Variable	Description
RS_SETUP_PROFILE	minimal, standard, or full (default: full)
JWT_RSA_PUBLIC_KEY_LOC	Load JWT public key from disk when .publickey index is not created

See docs/env-vars.md for full setup profile documentation.

---

Build & release

• Default version bumped from 9.3.0 → 9.4.0 (Makefile, Dockerfile).

---

Upgrade notes

• Existing deployments — No action required. Unset RS_SETUP_PROFILE keeps full behavior identical to 9.3.0.
• New lean / demo deployments — Set RS_SETUP_PROFILE=minimal when pointing at a shared or free-tier OpenSearch cluster (e.g. Aiven + Render). Only 4 meta indices are created; features like analytics, logs, rules, cache, and node tracking are unavailable until you move to standard or full.
• Health endpoints — /arc/health (upstream cluster ping, no auth) and /arc/_health (ARC node health) both work on minimal; _health returns node_count: 0 since .nodes is not created.

9.3.0

Summary

Feature release adding Elasticsearch Serverless support, including API key upstream authentication, automatic meta index remapping, Serverless-compatible index operations, and client-side index rollover. Also bumps the default version to 9.3.0 and updates the Go build image to 1.26.4.

---

New features

Elasticsearch Serverless support

ReactiveSearch API can now run against Elasticsearch Serverless clusters.

• Automatic Serverless detection — Detects Serverless via version.build_flavor from GET / and caches the result for the process lifetime.
• Meta index remapping — Dot-prefixed internal indices (e.g. .pipelines, .users, .analytics, .logs) are automatically rewritten to rs_* names (e.g. rs_pipelines, rs_users), since Serverless rejects user-created hidden indices. All plugins that use meta indices are updated to resolve names through this layer.
• Configurable prefix — Set RS_META_INDEX_PREFIX to force an alternate prefix on any cluster (must not start with _, -, or +). When unset, remapping applies only on Serverless.
• API key upstream auth — Set ES_API_KEY to authenticate to upstream Elasticsearch with Authorization: ApiKey … (recommended for Serverless). ES_CLUSTER_URL must not contain embedded credentials when ES_API_KEY is set. Basic auth via URL credentials remains supported for self-managed clusters.
• Serverless-compatible index creation — Strips platform-managed settings from index creation bodies on Serverless (index.hidden, shard/replica counts, auto_expand_replicas, etc.) via AdaptIndexBody.
• Index size via _cat/indices — Reindex and rollover logic uses _cat/indices for store size where _stats is unavailable on Serverless. Adds ParseStoreSize for human-readable size strings (1.2kb, 100mb, etc.).
• Client-side index rollover — Serverless rejects conditional rollover (max_age, max_docs, max_size) in the API body. On Serverless, ReactiveSearch evaluates those thresholds client-side and calls rollover without conditions when any threshold is met (OR semantics). Applies to analytics, logs, pipeline logs, and pipeline invocations. Cron schedules (@midnight, @hourly) and post-rollover cleanup (retain latest two backing indices) are unchanged.

Rollover thresholds (Serverless defaults)

Meta index	max_age (non-production)	max_age (production)
rs_analytics	30 days	30 days
rs_logs	7 days	30 days
rs_pipeline_logs	7 days	30 days
rs_pipeline_invocations	3 days	7 days

---

Configuration

Variable	Description
ES_API_KEY	API key for upstream Elasticsearch auth (e.g. Serverless). Mutually exclusive with credentials in ES_CLUSTER_URL.
RS_META_INDEX_PREFIX	Force meta index prefix (e.g. rs_) on any cluster; auto-applied on Serverless when unset.

See docs/env-vars.md for full details on meta index naming and Serverless rollover behavior.

---

Build & release

• Default version bumped from 9.2.0 → 9.3.0 (Makefile, Dockerfile).
• Go build image updated from golang:1.26.0 → golang:1.26.4.
• Binary publish workflow migrated to build from the OSS reactivesearch-api repo (drops confidential rs-api-server clone; packages build/ artifacts only).

9.2.0

Summary

Maintenance release with expanded GPT-5.2 and o-series model support, GitHub Actions model update, and Go base image upgrade.

Enhancements

• Adds support for new OpenAI models released after GPT-5.1:
  • GPT-5.2 family: gpt-5.2, gpt-5.2-pro, gpt-5-pro with 400K context window
  • o-series reasoning models: o1, o3, o3-mini, o3-pro, o4-mini with 200K context window
• Updates Code Review GPT workflow to use gpt-5 model

Chore

• Update base build to use Go 1.26.0, bringing performance and security improvements

Upgrade notes

• New GPT-5.2 models (gpt-5.2, gpt-5.2-pro, gpt-5-pro) are now supported with 400K token context windows. If you're using AI Answer or other OpenAI-powered features, these models are available alongside the existing GPT-5 family.
• The o-series reasoning models (o1, o3, o3-mini, o3-pro, o4-mini) are now also supported with 200K token context windows. Note: OpenAI considers o3 and o4-mini as succeeded by gpt-5 and gpt-5-mini respectively, but they remain available in the API.

9.1.0

Summary

Maintenance release with GPT-5 model support, Docker base image update, and suggestion system refactoring.

Enhancements

• Adds support for GPT-5 models including gpt-5, gpt-5-mini, gpt-5-nano, and gpt-5.1 with 400K context window

Refactoring

• Remove Zinc dependency from FAQ, Popular, and Featured suggestions

Chore

• Update base build to use Go 1.25.5, bringing performance and security improvements

Upgrade notes

• GPT-5 family models are now supported with their full 400K token context windows. If you're using AI Answer or other OpenAI-powered features, you can now specify gpt-5, gpt-5-mini, gpt-5-nano, or gpt-5.1 as your model.

9.0.0

Summary

Major release with pipeline scalability improvements, OpenAI model updates, and Elasticsearch v9 compatibility.

Enhancements

• feat(pipeline): add rollover for pipeline invocations (Jul 24, 2025)
• change(pipeline): allow up to 1000 env vars (Jun 13, 2025)
• update(openai): expand model support, including GPT‑5 and modern GPT‑4 variants (Sep 8, 2025)

Fixes

• fix: add compatibility layer for range queries to work with Elasticsearch v9 (Aug 29, 2025)
• fix(CI): run tests from module root to resolve internal package imports (Sep 8, 2025)
• fix(ES v9): add escompat file for range query (Sep 8, 2025)

Chore

• chore(build): increment release version (Jun 13, 2025)
• chore(env): use REACTIVESEARCH_ID over APPBASE_ID (Aug 29, 2025)
• chore(update): 9.0.0 (Sep 8, 2025)
• chore(code-review): use gpt-5 model (Sep 8, 2025)
• Merge branch updates into v9 and merge PR #167 (Sep 8, 2025)

Upgrade notes

• Pipelines using environment variables can now configure up to 1000 entries.

8.24.2

Fixes

• Dashboard analytics queries for pipeline when system index mapping may not be correctly set.

8.24.1

Changes

• feat(vector search): inherit value for applying vector search when vectorDataField is specified but value or queryVector isn't set in the component. This is a common scenario with ReactiveSearch UI kit where SearchBox (for suggestions or as input for kNN searching) and ReactiveList (vector search) are used in tandem.
• fix (suggestions): payload parsing where object is received

{
    "query": [
        {
            "id": "term",
            "type": "term",
            "dataField": "tags.keyword",
            "value": "Fashion",
            "execute": false
        },
        {
            "id": "one-line-search",
            "type": "search",
            "dataField": "one_liner",
            "value": "dresses",
            "execute": false
        },
        {
            "id": "search",
            "vectorDataField": "vector_data", // since no value or queryVector is provided, will get embedding for dresses and apply kNN search against vector_data field. Requires configuration of AI preferences for this to work
            "includeFields": [
                "name",
                "one_liner",
                "tags"
            ],
            "react": {
                "and": [
                    "term",
                    "one-line-search"
                ]
            }
        }
    ],
    "settings": {
        "useCache": false
    }
}

8.24.0

Enhancements

• Adds support for approximate KNN (ANN) for Elasticsearch and OpenSearch in ReactiveSearch API. The same search endpoint allows faceting and text filtering alongside kNN vector search. Note: For OpenSearch, ANN requires lucene or faiss engines for use with faceting and text filtering.
• When AI preferences are configured, setting queryVector in the API is optional

Example with ReactiveSearch API:

{
    "query": [
        {
            "id": "term",
            "type": "term",
            "dataField": "tags.keyword",
            "value": "Fashion",
            "execute": false
        },
        {
            "id": "one-line-search",
            "type": "search",
            "dataField": "one_liner",
            "value": "dresses",
            "execute": false
        },
        {
            "id": "search",
            "vectorDataField": "vector_data", // instead of "dataField" to indicate kNN search type
            "includeFields": [
                "name",
                "one_liner",
                "tags"
            ],
            "react": {
                "and": [
                    "term",
                    "one-line-search"
                ]
            },
            "value": "personalise"
        }
    ],
    "settings": {
        "useCache": false
    }
}

Using KNN when AI preferences are configured is as simple as setting vectorDataField instead of dataField. ReactiveSearch uses the AI preferences (embedding model configured, currently supports OpenAI) to generate a query vector from the user passed value. If AI preferences are not configured, queryVector is expected to be passed alongside.

Chore

• Updates to Go 1.23.5 (latest current version)
• Removes use of Sentry for remote error tracking

8.23.2

Fixes

• Auth headers handling to strip RS auth when querying ES endpoints

8.23.1

Fixes

• Updates synonyms index to avoid conflict with Elasticsearch's internal index.