Timeplus

Editor’s note: This Timeplus review is based on my hands-on evaluation of Proton 3.0.31 (open source) and Enterprise 3.4.1 in October 2026. I weighed pricing, setup, streaming SQL, connectors, paid features and rivals.

Quick verdict: I recommend Timeplus for teams that want streaming SQL over Kafka, Redpanda or Pulsar without a JVM stack, with live and historical queries in one engine. The trade-off: clustering and mutable streams are Enterprise-only, self-hosted pricing is by quote, and the community is small (about 2,300 GitHub stars).

Proton runs streaming and historical queries in one SQL dialect from a single C++ binary under 500MB, built on ClickHouse. Timeplus Inc., based in Santa Clara, open sourced it on September 21, 2023.

In this review, I’ll take a closer look at Timeplus’s pricing, features, and the Proton vs Enterprise split, so you can see whether it fits your pipeline.

Key Takeaways:

• Proton is free under Apache 2.0 with windows, stream joins, materialized views, Python and JavaScript UDFs, and native Iceberg.
• Cloud starts at $199 a month, capped at 32 CPU, with a 14-day trial.
• Enterprise is quote-only: clustering, mutable streams and the web console need it.
• No RIGHT or CROSS joins, and Kafka sink and Iceberg write issues remain open.
• Releases are frequent: Enterprise 3.4.1 (October 2, 2026) brought the Tabby agent; Proton 3.0.31 shipped September 17.

Pros and Cons

How Much Does Timeplus Cost?

Proton costs $0 under Apache 2.0, Timeplus Cloud starts at $199 a month, and self-hosted or BYOC Enterprise is a custom quote. Streaming SQL itself does not require Enterprise.

• Timeplus Proton (free, Apache 2.0): for single-node deployments via binary, Docker, Homebrew or Helm. No sales contact; community support on GitHub.
• Timeplus Cloud (from $199 a month): for teams that want managed, dedicated resources on a monthly subscription. It covers 1 to 32 CPU, 4GB to 128GB of memory and storage from 250GB SSD, with daily backups (2 versions kept), up to 20 team members, a 99.5% uptime SLA and support within 1 business day.
• Timeplus Enterprise self-hosted (custom quote): for multi-node clusters on Kubernetes or bare metal. Annual subscription billed monthly, with unlimited users, onboarding and training, a custom uptime SLA and support within 4 hours. Charges follow data processed and stored plus compute, and licenses are sized in CPU cores and GB of usage.
• Timeplus Enterprise BYOC on AWS (quote or AWS metering): for Enterprise inside your own AWS account on EKS. Bring your own license, or pay hourly via AWS Marketplace metering, where 1 MPU equals 1 deployed core. Sizes run from 4 MPU (16GB) to 32 MPU (128GB).

Payment options differ by edition:

• Cloud: credit card or AWS Marketplace.
• Enterprise: credit card, AWS Marketplace or purchase order.

A Startup Program targets pre-revenue and early-stage companies; its terms are not published, so ask before you sign.

Edition	Price	Trial	Scale	Support
Proton	$0 (Apache 2.0)	Not needed	Single node	GitHub community
Cloud	From $199/month	14 days, no card	1 to 32 CPU; 4GB to 128GB memory	Within 1 business day
Enterprise self-hosted	Custom quote	30 days, no card	Multi-node cluster, custom sizing	Within 4 hours
Enterprise BYOC (AWS)	License or hourly MPU metering	30 days, no card	EKS; 4, 8, 16 or 32 MPU	Within 4 hours

For context, RisingWave Cloud starts at $0.227 per RWU-hour with a 7-day trial, Materialize charges $1.50 per compute credit, and ClickHouse Cloud gives $300 of credit over a 30-day trial. Those billing units do not convert into each other, so compare them on your own workload.

Is Timeplus Good Value for Money?

• Good fit: teams that can run Proton single-node for $0, or that want a managed $199 starting point with an SLA.
• Less convincing fit: workloads that need more than 32 CPU on Cloud, or HA on a fixed budget, because both push you to a quote.
• Budget priority: stay on Proton and move to Enterprise only when clustering, mutable streams or the console justify the spend.

Reviewer’s Notes: My plan for any buyer: prototype on Proton, then run the same SQL through the 30-day Enterprise trial, and only then ask for a quote sized in CPU cores and GB. That order puts a real workload behind the number you negotiate.

My Experience With Timeplus

Timeplus has two front doors: a one-line Proton install that gives you a SQL engine in a terminal, and a one-line Enterprise install that opens a web console and starts the 30-day trial. Where Apache Flink needs a JVM cluster with a JobManager and TaskManagers before the first query, Proton starts from one binary.

Installing Timeplus Proton

I installed Proton with curl https://install.timeplus.com/oss | sh. The same engine is available three other ways:

• Homebrew: brew tap timeplus-io/timeplus, then brew install proton.
• Docker: the d.timeplus.com/timeplus-io/proton:latest image, exposing ports 8123 and 8463.
• Kubernetes: a Helm chart, still limited to a single node.

Running proton server started a config-less server that keeps its data in ./proton-data. From a second terminal, proton client connected me to the SQL prompt.

For one-off queries, proton local skips the server and client pair. Evaluate a pipeline in server mode instead, because a streaming query only shows its value while it keeps running against new events.

Starting the Timeplus Enterprise Trial

The Enterprise installer, curl https://install.timeplus.com | sh, downloads, unpacks and starts the server on Linux or macOS. The web console then lives at http://localhost:8000, and creating the first account starts the 30-day trial.

From there, my first session followed four steps:

1. Open Data Collection Wizard > Demo Stream for sample data, or connect Kafka, Confluent Cloud or Pulsar.
2. Run SELECT * FROM <stream> against that stream.
3. Open the Visualization tab to chart the query.
4. Pick Save As > Sink to send results to an output.

Sink destinations include:

• Messaging: Kafka or Pulsar.
• Analytics: ClickHouse or another Timeplus instance.
• Notifications: a webhook or Slack.

Enterprise runs on:

• Linux x64 or ARM: Ubuntu 20.04+, RHEL 8+, Fedora 35+ or Amazon Linux 2023.
• macOS: versions 14 and 15, on Intel or Apple silicon.
• Kubernetes: 1.25+ with Helm 3.12+.
• Windows: Docker only.

Where Proton Stops

Proton has no web console, no clustering and no mutable streams. Everything runs through the SQL client, so the demo stream, Visualization tab and Save As > Sink flow above are Enterprise only.

That matters if you demo the console to teammates during the trial: once the 30 days end, the UI they liked needs a paid Enterprise license, while the SQL engine underneath stays free in Proton.

Reviewer’s Notes: Start with the Enterprise trial if your team wants a UI for exploring streams and building sinks. Start with Proton if all you need is a SQL engine sitting next to Kafka.

Streaming SQL: Windows, Joins and Hybrid Queries

One Timeplus query can read a live stream, its stored history, or both, giving you streaming, historical and hybrid modes in one SQL dialect. Incoming events land in a write-ahead log first and are then flushed to ClickHouse-based historical storage.

That design suits pipelines that need both continuous calculations and historical backfills, such as real-time AI/ML features built with windows and materialized views. Flink SQL, by contrast, runs as a processing framework, not a database with its own storage.

The streaming SQL surface in Proton includes:

• Tumble, hop and session windows: fixed, overlapping and activity-based grouping.
• Watermarks: EMIT AFTER WINDOW CLOSE WITH DELAY handles out-of-order events, for example a 5-second tumble window with a 2-second delay.
• Stream joins: INNER by default, plus LEFT and FULL/OUTER, with range, bidirectional and enrichment variants.
• Incremental materialized views: results update as events arrive instead of being recomputed.
• Stream types: append, versioned and changelog streams.
• Dictionaries: lookup data for enrichment.
• Scheduled tasks: SQL that runs on a schedule next to continuous queries.
• JIT compilation: included in the free engine, not held back for Enterprise.

Treat watermark settings as a migration requirement. Events that arrive after the window closes plus the delay are dropped, so a 2-second delay has to match how late your producers actually deliver.

The gaps belong on your migration checklist:

• RIGHT and CROSS joins: not supported in the current version.
• Native RANK(): an open feature request since 2024.
• Nested UDFs: requested by users in 2024.
• Filling empty windows in streaming aggregation: open since 2023.
• sequenceMatch on streams: requested in 2024.

Port your hardest join and window first, not a simple filter that any streaming engine handles. A pipeline built on RIGHT joins or empty-window filling will not move over unchanged.

RisingWave speaks the PostgreSQL dialect, which wins when existing Postgres tooling drives the choice. Timeplus results are streaming-eventual rather than strict-serializable, so reconciliation work that must match a batch pipeline exactly belongs on Materialize.

• Best for: windowed analytics, enrichment and backfills that benefit from one engine for live and stored data.
• Skip if: you depend on RIGHT or CROSS joins, empty-window filling, or strict-serializable results.

Connectors, Iceberg, UDFs and Alerts

Proton and Enterprise share the same external-system list, so you do not pay just to connect Kafka or Iceberg.

The documented integrations are:

• Event systems: Kafka, Confluent Cloud, Redpanda and Pulsar.
• Databases: MySQL, PostgreSQL, MongoDB and ClickHouse.
• Lakehouse and storage: Apache Iceberg, S3, BigQuery and Databricks.
• Operational services: HTTP, Datadog, Elasticsearch and Slack.
• Other sources: log streams and remote Timeplus or Proton instances.

Enterprise 3.2 added NATS JetStream as a source and sink on top of that list.

Iceberg Without a Java Runtime

Timeplus reads and writes Iceberg through a native C++ implementation with no Java runtime, using an Iceberg REST catalog. Supported setups include the AWS Glue REST endpoint and Apache Gravitino, with S3 storage.

The catch: partitioned Iceberg writes have been an open request since December 2025. Check this before you point Timeplus at an existing partitioned lakehouse table.

UDFs and SQL-Defined Alerts

Timeplus supports 4 UDF types, not just JavaScript:

UDF option	Runtime or interface	Function types
JavaScript	Embedded V8	Scalar and aggregate
Python	Embedded CPython	Scalar and aggregate
SQL	SQL expressions	SQL-defined functions
Remote	HTTP	Externally executed functions

Enterprise 3.3.1 moved Python UDFs to free-threaded CPython 3.14 and stopped bundling third-party packages, so upgraders must reinstall them. Since 3.2.1, admins can also disable JavaScript UDF creation in the server config.

CREATE ALERT runs a streaming query that calls a Python UDF when conditions match, for example to post to Slack or send an email. Combined with scheduled tasks, it covers telemetry alerting without a separate rules engine.

Users have reported connector problems in the project’s GitHub issues:

• Protobuf imports: a Kafka schema registry URL does not resolve Protobuf import statements (opened December 2025).
• Kafka sink errors: a fatal-error thread has been open since October 2023, alongside a 2024 report of a DWAL_FATAL write error.

Turn those into acceptance tests: run your schema imports and sink failure recovery before you commit an operational workflow.

Kafka Connect plus ksqlDB covers similar ground but keeps you tied to the Kafka and Confluent stack. Timeplus is the better fit when Iceberg, ClickHouse and embedded Python need to live in one engine.

Timeplus Enterprise: Clusters, Mutable Streams and the Tabby Agent

Enterprise adds clustering for HA and horizontal scale. Every node runs the single timeplusd binary, with metadata, data and compute roles coordinated by NativeLog, a Multi-Raft replicated write-ahead log.

The paid additions are:

• Materialized view scaling and HA: continuous calculations spread across nodes with high availability.
• Web console: data exploration, resource management, pipeline lineage and observability.
• Timeplus Console: system metrics visualization.
• Mutable streams: primary-key tables that update in place.
• SLA-backed support: contracted response times instead of GitHub threads.

Mutable streams behave like a streaming MySQL or PostgreSQL table: a required primary key, row encoding, sharding by key, secondary indexes, TTL, and versioned or coalesced variants. They serve changelog processing and lookup data in joins.

Pay for Enterprise when mutable lookup state or HA is the requirement, not because a pipeline reads from Kafka.

What Shipped in 2026

• Enterprise 3.2: NATS JetStream and improved Okta SSO. Timeplus’s own figures claim up to 30x faster replication networking and up to 40x faster Kafka consumption for Protobuf and CSV.
• Enterprise 3.3: stream backup and recovery, plus materialized view rebalancing across the cluster.
• Enterprise 3.4: the Tabby data agent, an App Framework and App Marketplace, and experimental vector-similarity and full-text indexes.

Tabby works with OpenAI, Anthropic or compatible endpoints, including Bedrock. It proposes DDL and runs it only after approval, which is the boundary to keep in production.

Enterprise 3.4.1 shipped October 2, 2026 with Helm chart 14.0.0, and the 3.2 line still got patch 3.2.15 in August.

Proton reached 3.0.31 on September 17, about 31 releases since 3.0.1 in October 2025. Its 2,272 GitHub stars and 112 forks mark a small community, and many of the most-discussed open issues date from 2023 and 2024.

RisingWave keeps state in S3 with stateless compute, a different elasticity model from Timeplus’s local-state cluster. Compare that before you buy Enterprise for elasticity alone.

How Does Timeplus Compare to Competitors?

Timeplus wins on unified live and historical SQL in a compact engine; Flink wins on ecosystem, RisingWave on open-source scale-out, Materialize on consistency, and ClickHouse on analytics over stored data.

Tool	Starting price	Model	Best at
Timeplus	Proton $0; Cloud from $199/month; Enterprise quote	C++ streaming SQL database built on ClickHouse	Live and historical SQL in one binary
Apache Flink	$0 open source; managed on Confluent Cloud per CFU-minute	JVM stream-processing framework	Large stateful pipelines and ecosystem breadth
RisingWave	$0 open source; Cloud from $0.227/RWU-hour, 7-day trial	PostgreSQL-compatible streaming database with S3 state	Postgres tooling and elastic scale-out
Materialize	$1.50/compute credit	Streaming database on differential dataflow	Strict-serializable incremental views
ClickHouse Cloud	Metered; $300 credit over a 30-day trial	Real-time column store	Analytical queries on stored data
ksqlDB on Confluent	Part of the Confluent platform	Kafka-native stream processing	Teams committed to Confluent

• Timeplus vs Apache Flink: Flink is the pick for large stateful pipelines and teams already comfortable running its JVM cluster. Proton removes that cluster for a single-node SQL workload, but a smaller binary does not replace Flink’s ecosystem or prove faster jobs.
• Timeplus vs RisingWave: RisingWave wins when PostgreSQL compatibility and open-source scale-out are requirements, since its S3-backed state separates storage from stateless compute while Timeplus puts clustering behind Enterprise. Timeplus stays ahead when you want one compact engine for live events plus retained history.
• Timeplus vs Materialize: Materialize is my choice for strict-serializable incremental views where reconciliation correctness drives the architecture. Timeplus’s windows, UDFs and historical queries suit operational analytics, but they are not a substitute for that consistency guarantee.
• Timeplus vs ClickHouse: ClickHouse wins for analytical queries on stored data when continuous processing is not the missing piece. Proton is built on ClickHouse and can pre-process streams before ClickHouse ingestion, so running both often beats replacing one.
• Timeplus vs ksqlDB: ksqlDB fits when Kafka and Confluent already define the platform. Timeplus is the stronger pick when the pipeline also needs Iceberg, embedded Python or historical SQL inside the processing engine.

For an existing Flink or ksqlDB deployment, make a job-by-job migration decision, not a platform switch based on install size. Unsupported joins, sink recovery behavior and paid HA can outweigh Proton’s single binary.

My shortlist is Proton plus the rival that matches your hardest requirement: RisingWave for Postgres and elastic state, Materialize for consistency, or Flink for complex stateful processing. Run the same job on both during Timeplus’s 30-day Enterprise trial or RisingWave’s 7-day trial, and let the disqualifying requirement decide.

How I Reviewed Timeplus

I worked through Proton 3.0.31 and Enterprise 3.4.1: the Proton install paths, and the Enterprise trial flow from account creation through the demo stream, queries, the Visualization tab and sinks.

I then went through the streaming SQL surface (windows, watermarks, joins and materialized views), the connector list, Iceberg support, the four UDF types and alerts, plus clustering, mutable streams and the 2026 release history.

I compared every pricing route Timeplus offers: Proton, Cloud, self-hosted Enterprise and AWS BYOC. Then I set Timeplus against Flink, RisingWave, Materialize, ClickHouse and ksqlDB. I make no throughput or latency measurements of my own; the 30x and 40x figures are Timeplus’s.

Prices current as of October 2026.

Should You Use Timeplus for Streaming SQL?

Timeplus suits small and midsize teams that want streaming and historical SQL without a JVM stack. Proton already covers windows, joins, UDFs and Iceberg, but production HA and mutable streams need an Enterprise budget.

• Kafka pre-processing into ClickHouse, real-time dashboards or alerts: start on Proton.
• HA, mutable streams or a UI: run the 30-day Enterprise trial, then get a quote.
• A managed service under 32 CPU: Cloud from $199 a month, after checking that 128GB of memory and the storage tier cover your retention.
• Deep Flink skills or very large state: stay on Flink.
• Postgres compatibility or S3-backed elastic state: choose RisingWave.
• Strict consistency: choose Materialize.

Test your hardest SQL and sink behavior before you pay for the console or clustering. The missing RIGHT and CROSS joins and the Kafka sink issues open since 2023 matter more to a migration than Tabby.

Start with Timeplus: install Proton with curl https://install.timeplus.com/oss | sh and port one existing streaming job before you talk to sales.

FAQs

Is Timeplus free?

Timeplus Proton is free under Apache 2.0 and needs no sales contact. Timeplus Cloud starts at $199 a month after a 14-day trial, and Enterprise offers a 30-day trial before a quoted license. Proton is enough when a single node handles your volume and you can live without the web console, clustering and mutable streams.

What is the difference between Timeplus Proton and Timeplus Enterprise?

Proton and Enterprise share the same SQL engine and connector list. Enterprise adds clustering and HA, materialized view scaling, mutable streams, the web console, Timeplus Console for metrics, and SLA-backed support. Windows, joins, materialized views and UDFs do not require the paid edition, so choose Enterprise for those operational needs only.

Can I self-host Timeplus Enterprise?

Yes, Enterprise is self-hostable on bare metal or on Kubernetes 1.25+ with Helm, and as BYOC inside your own AWS account on EKS. Linux and macOS install natively, while Windows works only through Docker. Get CPU, memory, storage and support terms into the quote before you commit to the annual subscription.

Does Timeplus support Apache Iceberg?

Yes, Timeplus reads and writes Iceberg natively through an Iceberg REST catalog, with no Java runtime. Supported setups include the AWS Glue REST endpoint and Apache Gravitino with S3 storage. Partitioned writes have been an open request since December 2025, so check that against your lakehouse layout before you adopt it.

Is Timeplus better than Flink?

Timeplus is lighter to run when you want one binary that combines streaming and historical SQL, with no JVM cluster. Flink has the bigger ecosystem and suits very large stateful jobs. For a new SQL-led pipeline, Timeplus is the simpler start; for established Flink jobs, check SQL compatibility and HA needs before you switch.