AlgoX2 manages the entire data plane the way an operating system manages hardware — so you focus on business logic, not infrastructure. Runs on-prem, in your cloud, or hybrid.
| Sub-10μs | 100M msg/s | 10× | 90% |
|---|---|---|---|
| Publish acknowledgements | Per-node throughput | Lower server count | TCO reduction |
What makes AlgoX2 different
-
Unified Data Plane: App-to-app, app-to-storage, and storage-to-app — all natively, in one engine. No connector sprawl. One data path from ingestion to lakehouse.
-
Partition-Free API: No Kafka-style topic partitions to size, manage, or rebalance. Producers publish to a single stream; X2 handles ordering and replication internally. 100K+ logical streams without file-descriptor explosion. Scale by adding nodes, not reshuffling data.
-
Internal Multicast Fan-Out: One write serves thousands of consumers. The network switch replicates between AlgoX2 nodes — no duplicate CPU or bandwidth work. 10 nodes × 100 Gbps = 1 Tbps of fan-out. Clients connect via standard TCP/TLS.
-
Compute–Storage Separation: CPU, network, and storage scale independently. Policy-driven tiering across NVMe and object storage. No over-provisioning.
-
EXEC — Code Inside the Stream: Run any program (Python, Rust, C++, Go) inside the cluster. Messages delivered via shared memory in microseconds. No external process fleets.
-
Deterministic Sequencing: Every event gets a globally unique sequence number. One canonical order. Recovery is replay, not re-coordination.
-
Multi-Protocol & Protocol Bridging: Kafka, NATS, MQTT, Redis, HTTP, WebSocket — all native. Publish via one protocol, subscribe via another. No connectors, no middleware, no translation layers.
Streams land as open tables by default
The same data that serves real-time consumers is automatically available for analytical queries — as Iceberg, Delta, or Hudi tables in S3-compatible Parquet storage. No ETL pipelines. No manual tier tuning.
Everything a streaming backbone needs
| Capability | Description |
|---|---|
| Inline Transforms | Filtering, format conversion, compression — deterministic and bounded. |
| Lakehouse Materialization | Streams land as Iceberg, Delta, or Hudi tables. S3-compatible Parquet. |
| Enterprise Security | Fine-grained ACLs, TLS, mTLS, quotas, audit logs. |
| Native Multi-Tenancy | Tenant namespaces with enforced policies, quotas, and isolation. |
| Zero Data Loss | Three copies on separate nodes. Append-only log. Transactional semantics. |
| Deploy Anywhere | On-prem, BYOC, hybrid. Single node to hundreds. Same binary everywhere. |
Order once. Execute everywhere.
Layer 1 — Sequencer: Every event receives a globally unique sequence number. Unordered inputs become one deterministic timeline.
Layer 2 — Execution: Any number of processes execute against that sequence independently and in parallel — synchronized through the journal, not through each other.