Timeplus Enterprise 3.4: From Idea to Production, Full Lifecycle Real-Time Pipelines Powered by Tabby Agent
When we launched Timeplus 3.0, we made a simple promise: real-time stateful pipeline that doesn’t make you choose between low latency and massive scale. With 3.3, we made that promise hold up in production. Today, many teams run network and security detection at millions events per second with sub-second latency, and gaming telemetry at PB of data scale, on one engine.
The questions we hear from teams have changed. They used to ask whether the engine could take the load. Now they ask: How fast can we build on it? Can we iterate quickly when the data, schema or the threat changes? How can we package and maintain applications? And do we have enough head room to keep the cost and the effort down as it grows?
That’s what our 3.4 release is about. Meet Tabby, the Timeplus Agent. Ask it a question and it explores your streams, helps write the SQL, build, test and monitor on your pipelines, all under your permissions. Package the whole pipeline, SQL and dashboards included, into an app you can install in one guided flow, and start from 10+ we ship ourselves, from cyber detection to OCSF to AI/ML feature stores.
Underneath, 3.4 introduces new vector and full-text indexes to support more real-time AI workloads, and hardens the platform with smart tiered storage, resource policy, and distributed query improvements, and many more platform enhancements.
By “agentic,” we don’t mean pipelines running off on their own. We mean an agent that helps you build and run them, and you observe and control how much it’s allowed to do. From idea to production, with a lot less hassle.
Meet Tabby, Your Timeplus Agent
“Timeplus’ Tabby Agent has already been useful. Today, I noticed that we had some issues in the pre-prod and I used the agent to debug the problem … In the end, everything worked out fine and the agent was really helpful to figure out both the issue and the solution.”
— Our Timeplus customer in the gaming industry
Developing a real-time pipeline involves translating requirements into SQL, validating results against incoming events, deploying changes, and investigating operational issues. Tabby, the Timeplus Data Agent, brings AI assistance to these tasks directly inside your workspace.
Consider a security analytics pipeline that detects traffic spikes in Cisco ASA firewall logs. The pipeline parses raw syslog messages, maintains per-source traffic baselines, and flags activity that exceeds a configured threshold. The Cisco ASA DDoS Detection demo includes simulated background traffic and a controllable attack stream, providing a repeatable way to validate detection behavior.
1. Design
Ask Tabby to inspect the Cisco ASA firewall log schemas and propose a pipeline that parses raw syslog, maintains per-source traffic baselines, and detects spikes against those baselines.
2. Build
Have Tabby help write the streaming SQL and materialized views for parsing logs, aggregating live traffic, calculating baselines, and identifying sources that exceed the configured spike threshold.
3. Validate
Use the demo’s controllable attack stream to generate a traffic burst. Ask Tabby to query the detection output and check whether the simulated attack is flagged, while examining how background traffic compares against its baselines.
4. Deploy
Review the proposed SQL and approve changes to the detection pipeline. Tabby executes statements under the signed-in user’s permissions, subject to the administrator’s configured approval mode.
5. Observe
Ask Tabby to inspect the pipeline for failures and lag, examine traffic rates and spike ratios, and investigate whether an unexpected result comes from incoming activity or a processing issue.
You control how Tabby operates. Administrators can choose read-only, approval-required, or unrestricted SQL modes. Every SQL statement runs as the signed-in user, with Timeplus’s existing role-based permissions enforcing access.
Tabby also supports durable memory across conversations and reusable skills for recurring workflows. Read-only background tasks can carry out standing instructions on a schedule or in response to new data, with results delivered to a dedicated conversation. For workflows that extend beyond Timeplus, Tabby can connect to HTTP-based MCP tool servers with per-server approval policies.
You choose the model, too. Tabby supports OpenAI, Anthropic, and compatible endpoints, including self-hosted options and Amazon Bedrock.
Apps Framework: Package, Share, and Deploy Your Pipeline Applications in Minutes
A useful real-time pipeline application usually includes several pieces: data connections, streams, transformations, materialized views, and dashboards. Bundling those pieces, maintaining, sharing and deploying these can be very daunting and takes time.
The new Timeplus App Framework packages them into a single installable .tpapp file.
From the built-in app catalog, teams can discover solutions for AI agent monitoring, network security, and continuously updated ML features. A guided installation flow collects the required configuration and deploys the bundled pipelines and dashboards together. Apps can also be installed directly from a package file.
Each installation gets a dedicated database, allowing separate instances of the same app to serve different data sources. App detail pages bring configuration, resources, lineage, dashboards, and lifecycle history into one place.
The framework also handles the ongoing lifecycle. Failed installations roll back, upgrades add resources and replace dashboards while preserving existing data, and uninstalling gives operators a choice to retain or delete the app’s data.
For developers and partners, the same framework provides a way to turn proven SQL workflows into reusable products. Package streams, materialized views, UDFs, and dashboards with a declarative manifest, then distribute the app privately or through the community app catalog.
A Stronger Platform for Versatile Real-Time Workloads
Timeplus 3.4 introduces new vector-similarity and full-text indexes for real-time AI use cases, and also improves how the core engine handles storage, memory, and recovery under sustained ingestion.
Search Capabilities: This release introduces vector-similarity and full-text indexes. Vector indexes accelerate nearest-neighbor searches over embeddings, opening a path to semantic retrieval inside Timeplus. Full-text indexes accelerate token searches over string data.
Both capabilities are available in 3.4. They remain experimental and are not yet recommended as the primary access path for production-critical queries.
Smarter tiered storage helps reduce small-object buildup in cold storage. Eligible parts can merge before moving, while a new grace period gives newly arrived, isolated parts time to find a merge partner. The result is fewer, larger objects in storage such as S3, reducing PUT requests and improving compression.
Configurable primary-key index memory limits give operators another way to manage resource usage. A new per-stream setting caps cached index memory, evicting less recently used indexes and reloading them when needed. This limit is opt-in; the default retains unlimited caching.
Distributed-query reliability address shard counting, node-targeted historical queries, replication recovery, checkpoint progress, failed historical commits, and mutable-stream behavior. Together, these changes strengthen the foundation that long-running streaming applications depend on.
Get Started with Timeplus 3.4
Timeplus 3.4 gives teams new ways to work with real-time data: ask Tabby to investigate a workspace, install a packaged solution, or turn an existing pipeline into an app others can reuse.
Tabby appears by default after upgrading, but makes no model calls until an administrator configures an LLM endpoint. Existing users should review the upgrade notes for changes affecting certain materialized views, mutable-stream indexes, and Pulsar TLS connections.
Download Timeplus, explore the 3.4 release notes, and review the upgrade guide to get started.
