Stop asking customers to troubleshoot your tickets ask your AI agent instead
Sending customers a checklist of commands and log requests from Kafka? Let AI agents inspect live data, logs, and data layout automatically and save the back and forth.
The missing operational context between knowledge and action.
Knowledge Agent
Glean · CoPilot · custom RAG
What do we know?
Operational Context Agent
Streambased
What is actually happening?
Action Agent
Claude · Confluent MCP · mcp-kafka
Fix it.
AI can draw from previous knowledge and help take actions. It still can't help investigate.
Most AI platforms can retrieve documentation or execute workflows. What they're missing is an understanding of what's happening right now in production.
Today: Knowledge
AI can answer from the docs
RAG over large documentation repositories — Glean, Guru, Confluence AI, internal wikis. Great at “how does feature X work?” and “what are the expected behaviours?”
Today: Actions
AI can execute operational tasks
Agent frameworks and MCP servers restart services, roll back deployments, scale infrastructure, open Jira tickets, and update configuration.
What's missing
An understanding of right now
Support teams already know how to investigate complex customer issues. What slows them down is gathering the right operational context: asking customers to run commands, collect logs, and explain what changed.
Streambased gives humans and AI agents direct access to the live data behind each ticket, helping them test hypotheses, identify anomalies, and reach the root cause faster.
Questions an agent can't answer today:
- ?What is happening right now?
- ?Is this normal?
- ?What changed?
- ?When did it begin?
- ?Has this happened before?
- ?Is this isolated or widespread?
- ?Which customers are affected?
- ?What is the likely root cause?
See it in action
Streambased MCP, live in Zendesk
Watch a support agent investigate a real issue without leaving their helpdesk, Streambased answering from live and historical operational data through MCP.
One continuous timeline
From the last millisecond to years of history
An agent doesn't think in terms of "the stream" and "the warehouse." It asks a question and reasons across time. Drag the handle from the live edge back through years of archive — the same query keeps working, and Streambased picks HOT, UNIFIED, or COLD automatically.
- HOT reads the live stream in milliseconds
- UNIFIED spans stream and archive in one result
- COLD reaches years into compacted history
mode = HOT
Reading the live stream
now
p50 4 ms
Auto-playing·drag, tap a marker, or use ← → to take control.
Continuous operational awareness, six ways
Instead of asking many systems independently, an agent reasons across the complete operational history of a business.
Continuous state querying
Retrieve the current operational state across every data source at once. “Did order X (from topic orders) get delivered (from topic deliveries)”
Temporal consistency
Compare live behaviour against historical baselines. Last minute vs 30-day average, today vs same day last week, this deployment vs the last one.
Operational reasoning
Move beyond querying into explanation. The service tells an agent what changed, why, and how confident it is, in deep detail, not just scratching the surface.
Cross-domain correlation
Reason across system logs, metrics, customer interactions, business events and much more in a single line of thought.
Historical investigation
Search years of operational history. Has this happened before? Which incidents looked similar? Which deployment introduced it?
Business context
Unlike observability tools, Streambased understands customers, merchants, orders, payments, subscriptions, devices, and regions as first-class entities.
Agents reason about your business, not your schemas
Streambased exposes a semantic operational model. Pick an entity to see the attributes it carries and the questions it unlocks.
Attributes
Questions an agent can ask
- →What is ACME's revenue doing versus last quarter?
- →Which plan tier is churning fastest in the EU?
- →Show this customer's orders in the last 24 hours.
Example workflow
"We are not receiving orders."
The operational context agent can determine where the blockage is, correlating live data with years of history to determine what has changed and then returns an explanation in one step, with no back and forth.
→The issue began three minutes after deployment 1843.
→Payment failures increased only for UK merchants using Gateway B.
→Overall traffic remains normal — this is isolated, not systemic.
→This matches an incident from April, resolved by rolling back a configuration change.
Recommended action
Rollback payment configuration.
Temporal intelligence
Is this normal? cannot be answered with 1 day's data.
A raw number means nothing on its own. Streambased compares live behaviour against its own historical baseline, so an agent knows when something has genuinely broken, when it started, and what happened immediately beforehand.
- Last minute vs 30-day average
- Today vs the same day last week
- This deployment vs the previous one
Checkout latency
last 60 min vs 30-day baseline
326 ms
+172% vs baseline
Not just "latency is 340 ms" — the agent knows it's 172% above the 30-day norm and broke three minutes after deployment 1843.
Where Streambased fits
The operational context layer that sits between knowledge retrieval and operational actions.
| RAG platforms | Observability | AI BI platforms | Streambased | |
|---|---|---|---|---|
| Documentation | — | — | — | |
| Live operational state | — | Limited | — | |
| Historical business data | — | — | ||
| Streaming events | — | Limited | — | |
| Cross-domain reasoning | — | Limited | Limited | |
| Temporal comparisons | Limited | Limited | ||
| Business + operational context | — | Limited | Limited | |
| AI operational reasoning | Limited | Limited | Limited |
Give your AI agents operational awareness
See how Streambased gives agents a continuously updated understanding of your business, from the last millisecond to years of history.