Let’s be honest: building real-time data pipelines has historically been a test of patience. Between searching through nested wizard menus, parsing endless adapter properties, and hand-crafting schema mappings, going from a high-level design to active streaming database replication can take a full working day – at least.
The alternative is to design, monitor, and troubleshoot the pipeline by having a conversation about it.
What is Clara, and where does it live in Striim?
Clara is a documentation-grounded conversational agent built into the Striim platform, reachable from the Ask Clara panel and the Explore Striim using AI box on the Home page. It builds pipelines, answers connector documentation questions, reports cluster health, and diagnoses failed applications. It translates prompts into production-grade streaming pipelines while keeping you in control.
What does Clara actually know?
Clara understands Striim’s entire connector catalog and can design virtually any replication, routing, or enrichment flow across hundreds of sources and targets:
- Database replication — PostgreSQL-to-BigQuery, Oracle-to-ClickHouse, and equivalent source/target pairs
- Warehouse and lakehouse ingestion — including Databricks
- Operational filtering — continuous queries that drop or route events in flight
- PII masking — transformation applied in the stream, before the data lands
What are the four jobs Clara does?
- Builds pipelines from a plain-language intent, in AI Mode
- Answers documentation questions about connectors and their properties
- Monitors the cluster and returns structured system stats
- Diagnoses crashed applications down to the root-cause exception
How do you choose between Initial Load and CDC?
Initial Load is the one-time, full extraction and migration of all historical data from a source system to a target destination. Change Data Capture is the continuous process of identifying and capturing only the subsequent inserts, updates, and deletes, keeping the target synchronized in real time. Clara asks which one you want before it configures anything.
| Initial Load | Change Data Capture (CDC) | |
|---|---|---|
| What it moves | All historical data in the source | Only subsequent inserts, updates, deletes |
| When it runs | Once | Continuously |
| Target state | A point-in-time snapshot | Synchronized in real time |
| Choose it when | You need a one-time migration of what already exists | You are powering real-time analytics and AI-powered data applications |
Clara handles this as a fork in the conversation, not a buried radio button: it offers Initial load only or Change data capture only and waits.
How does the conversational build flow actually work?
You type your intent into the Explore Striim using AI box and Clara runs the build in AI Mode. It confirms the application name and namespace, asks whether you want Initial Load or CDC, opens the Configure Oracle Source form for visual table selection, compiles the TQL, and wires the components onto the interactive canvas once you approve.
The five steps, end to end
Start from the Home page and enter your intent:
“I want to build a real-time CDC pipeline from Oracle to ClickHouse to power real-time analytics and AI-powered data applications.”
- State the intent. One sentence, in the Explore Striim using AI box.
- Confirm naming and strategy. Clara verifies naming availability and returns the fully qualified name (admin.OracleToClickHouseApp), then asks how you want to read your data — Initial Load or CDC.
- Select the source tables visually. Clara opens the Configure Oracle Source form right in the UI so you pick specific tables (EMPLOYEES, DEPT_EMP, DEPT_MANAGER) instead of capturing entire schemas.
- Review the generated TQL plan. Nothing has been deployed yet.
- Deploy and start the app. Clara wires the components across the canvas and launches the active stream.
What does the generated TQL contain?
The plan is a complete, readable application definition, not a black box. For the Oracle-to-ClickHouse example it includes:
- CREATE SOURCE … USING OracleReader with Username, Password, the JDBC URL, the explicit Tables list, and Mode: ‘CDC’
- OUTPUT TO admin.OracleToClickhouseSync_Stream
- CREATE TARGET … USING ClickHouseWriter with connectionProfileName, the matching Tables mapping, and TableEngine: ‘MergeTree’
- Write behavior: Mode: ‘APPENDONLY’, DeleteStrategy: ‘LightweightDelete’, UploadPolicy: ‘eventcount:10000,interval:30s’, CDDLAction: ‘Process’, NullMarker: ‘NULL’
- INPUT FROM admin.OracleToClickhouseSync_Stream and END APPLICATION
What stops the agent from deploying something you didn’t approve?
Enterprise pipelines require absolute safety, so Clara compiles your selections into a complete TQL plan and then stops. No schema changes are deployed and no code is run until you review and explicitly approve the generated plan. You inspect configurations, continuous queries, and destination credentials in the workspace first.
What you can inspect before anything runs
- Pipeline configuration — every source and target property, as written
- Continuous queries — real-time filtering or transformation logic
- Destination credentials — connection profile and target details
- Connectivity — Validate Connections tests before you commit
Your three choices on the plan are Discard, Validate Connections, and Accept TQL. Only after Accept TQL does Deploy and Start this App become the next move.
Can Clara answer documentation questions and monitor the cluster?
Yes. Beyond building pipelines, Clara sits alongside you as a documentation expert and an active cluster operator, handling technical lookup fatigue and environment auditing. Ask it for connector properties and it returns mandatory properties, syntax examples, and critical warnings. Ask it for system stats and it returns structured tables of CPU rate, memory, agents, and lag.
Documentation audits
Direct questions get direct answers, instantly:
- “Give me all properties of OracleReader”
- “How do I configure supplemental logging?”
Cluster monitoring and commands
Ask “Give me my current system stats” or “What’s the current lag on my OracleIngestion app?” and Clara returns structured tables rather than prose:
| Table | Fields returned |
|---|---|
| Server Node Stats | Node Name, CPU Rate, Free Memory, Uptime, Version |
| Storage & Throughput Stats | Cluster Storage Free, Cluster Storage Total, Receive Throughput, Transmit Throughput |
| Application Stats | Application Name, Status, Servers, Processing Rate, Source Rate, Latest Activity |
In practice that means reading a CPU rate of 18% utilization, free memory in GB, active agents, and lag as a table you can act on. You can also deploy apps via direct commands from the same panel.
How does Clara troubleshoot a pipeline that crashed?
Ask it. Data pipelines run in dynamic environments where a mismatched host, an expired credential, or a formatting error halts a deployment. Instead of parsing a multi-page stack trace, ask “Why did OracleEmbeddingPipeline crash? Walk me through the error and how to fix it.” Clara parses the Java exception, explains the offending property, and returns documented remediation.
The three things Clara does with a failure
- Pinpoints the failure. It parses the Java exception to find the root cause.
- Explains the “why.” It identifies the property that caused the issue.
- Provides remediation. It returns the documentation steps that solve it.
A live diagnostic walkthrough
For a terminated admin.OracleToClickhouseSync, the analysis comes back as a structured root-cause report:
- Failed component: admin.OracleToClickhouseSync_Source (SOURCE)
- Exception: com.striim.exception.checked.AdapterInternalException
- Error message: No suitable driver found for …
- Explanation: the source adapter failed to initialize because Java’s JDBC DriverManager could not locate an appropriate driver for the connection configuration provided
That error typically occurs under one of two conditions:
- Missing JDBC driver — the Oracle JDBC driver JAR (e.g. ojdbc8.jar) is not present in the Striim server classpath ($STRIIM_HOME/lib or $STRIIM_HOME/setlib)
- Missing or invalid connection URL — the Connection URL property is blank, incomplete, or missing the expected JDBC prefix (e.g. jdbc:oracle:thin:@…)
The recommended resolution is to check the source’s settings so the Connection URL is fully specified (jdbc:oracle:thin:@//<host>:<port>/<service_name>) and confirm that host, port, user, and password are properly configured.
How do you run Clara inside your own network?
Two paths. Connect directly to Striim’s cloud demo servers with zero configuration to evaluate Clara immediately, or deploy Clara inside your own enterprise network boundary – locally via Docker or on a remote server – using your own OpenAI or Gemini API keys to maintain complete compliance and data privacy.
| Striim-hosted demo servers | Within your own network | |
|---|---|---|
| Setup | Zero configuration | Docker locally, or a remote server |
| Where it runs | Striim’s cloud | Inside your enterprise network boundary |
| LLM keys | Provided | Your own OpenAI or Gemini API keys |
| Best for | Evaluating Clara’s capabilities immediately | Complete compliance and data privacy |
How to apply this: a checklist
- Open the Ask Clara panel and state the pipeline you want in one sentence, naming source and target.
- Decide Initial Load vs. CDC before you build – one-time historical migration, or continuous sync.
- Select tables explicitly in the Configure Source form rather than capturing whole schemas.
- Read the generated TQL line by line, especially Mode, UploadPolicy, and DeleteStrategy.
- Run Validate Connections before you hit Accept TQL.
- Deploy and start the app, then confirm it on the canvas and in Application Stats.
- Ask Clara for system stats on a schedule instead of hunting through monitoring screens.
- Send crashes straight to Clara with the app name and ask for root cause plus remediation.
Disclaimer: Clara is an evolving collaborative AI assistant. Always review generated TQL plans, diagnostic scripts, and configuration recommendations before deploying changes to live production environments.
Ready to build real-time architecture with Clara? Reach out to our team, or get started with Striim for free, today.
