risingwavelabs /risingwave
Event streaming platform for agentic AI. Continuously ingest, transform, and serve event streams in real time, at scale.
★ 9,413⑂ 866👁 80⑃ 19,376● 8,478▤ 129
+1
日增长
9,413
stars
866
forks
80
watchers
19,376
prs
8,478
issues
129
releases
README
🌊 Event Streaming for Agentic AI
Docs | Benchmarks | Demos | Case Studies
RisingWave is an event streaming platform for agentic AI. It continuously ingests data from databases, event streams, and webhooks, processes it incrementally, and serves fresh results at low latency, replacing the traditional event streaming stack (e.g., Debezium + Kafka + Flink + serving DB) with a single system.
!RisingWave
---
Try it out in 60 seconds
shell
curl -L https://risingwave.com/sh | sh
For Docker, Kubernetes, and other options, see the quick start guide.
---
The problem
Agents and real-time applications need data that is always fresh and queryable at low latency. The standard approach chains together Debezium for CDC, Kafka for transport, Flink for processing, and a database for serving. Each hop adds latency and each system adds operational overhead.
RisingWave replaces the whole stack: ingest, process, serve, store.
---
How it works
Ingest from any source
RisingWave ingests across the full data spectrum:
- Webhooks: HTTP-based event ingestion from SaaS applications and external systems
- Database changes: native CDC from PostgreSQL, MySQL, and others via transaction log reading
- Event streams: Kafka, Pulsar, Kinesis, and other message brokers
- Historical data: batch ingestion from S3, data warehouses, and other storage systems
All sources are unified under the same SQL interface. Streams and tables can be joined freely.
Process continuously
RisingWave performs incremental computation over ingested data. When upstream data changes, only the affected results are recomputed. End-to-end freshness is under 100 ms.
This is the core mechanism behind everything RisingWave does: materialized views that are always up to date, without full recomputation on every query.
Serve at low latency
Query results are maintained in RisingWave's internal row store and served at 10-20 ms p99 latency. Agents and applications query this layer directly using standard SQL. No polling, no cache warming, no TTL management.
Store in Apache Iceberg™
For long-term retention and analytical access, RisingWave writes to Apache Iceberg™ tables. It hosts the Iceberg REST catalog directly and handles table maintenance — compaction, small-file optimization, snapshot cleanup — without external tooling. Iceberg queries are executed via Apache DataFusion, a vectorized query engine. Because Iceberg is an open format, data is also readable by Spark, Trino, DuckDB, and other engines.
The row store and Iceberg layer serve different purposes: the row store is for low-latency serving, Iceberg is for durable, open-format storage and analytical queries. RisingWave manages both.
---
Use cases
- Monitoring and alerting: continuous evaluation of streaming metrics against thresholds
- Feature stores: batch and streaming features computed over the same pipeline, served from the same system
- Live dashboards: materialized views updatTop issues
#15963Persistent Storage (S3) grows indefinitely25 comments · 2024-03-27#2897feat(executor): Stateful operators should stream newly arriving messages24 comments · 2022-05-30#17296panic when using Iceberg Sink connector from materialized view22 comments · 2024-06-18#8712perf: nexmark q022 comments · 2023-03-22#7343perf: nexmark q521 comments · 2023-01-12#5214Imbalanced kafka source actors' throughput when running nexmark benchmark21 comments · 2022-09-08#24673foyer: error handling improvement19 comments · 2026-02-02#17451the auto schema detection does not order the columns in the same way for different ables19 comments · 2024-06-25
Discussions
#27552How can I tell when a decoupled S3 sink has finished processing a batch?0 comments · 2026-10-19#26850RisingWave into Apache Arrow through ADBC, using psqlodbc 16 (adbcBridge 0.1.0)1 comments · 2026-08-29#26533Secrets hygiene in S3 e2e test helpers (see #26532) — is this worth fixing?0 comments · 2026-08-01#25229how to renew lakekeeper iceberg catalog token0 comments · 2026-04-02#25176what consumer.ack_policy should be used for a nats source ?1 comments · 2026-03-25#24458When building a materialized view based on an Iceberg external table, should it be treated as a source or a table?0 comments · 2026-01-14#24321Illegal instruction (SIGILL) when running RisingWave on CPU without AVX2 (testing only)0 comments · 2026-01-02#24156Metabase UI Visibility Depends on pg_catalog Permissions in RisingWave0 comments · 2025-12-17
Latest release
v3.1.1 v3.1.1 2026-11-05release v3.1.1 What's Changed fix(frontend): enforce source SELECT privilege for postgres_query (26862) by @risingwave-ci in https://github.com/risingwavelabs/risingwave/pull/27218 cherry-pick fix(meta): preserve connection and secret refs during connector alters (26923) to release-3.1 by @maybe-vibe in https://github.com/risingwavelabs/risingwave/pull/27200 cherry-pick: feat(stream): support keyed changelog planning and execution (27184, 27206) by @howenyap in https://github.com/risingwavelabs/risingwave/pull/27252 feat(stream): support broadcast temporal join (26932, 27025, 27000