Overview
Your debugging partner that investigates production issues using natural language conversations instead of dashboards.
Sazabi Chat is a chat-first interface for investigating production issues. Instead of clicking through dashboards and writing queries, you describe what you want to know in plain language and the assistant finds the answers.
What is Chat?
Chat is your debugging partner. It has access to your logs, traces, and metrics, and can search, correlate, and analyze data on your behalf. When something breaks in production, you open a conversation and describe the problem. The assistant investigates, gathers evidence, and reports back with what it found.
Think of it as having a senior engineer who knows your entire system, available 24/7, ready to help you investigate any issue.
The chat interface
The chat interface consists of several key components:
- Composer: Where you type your questions and messages. Supports markdown formatting and can include code snippets.
- Messages: The conversation history between you and the assistant. Your messages appear on the right, the assistant's responses on the left.
- Tool calls: When the assistant needs to search logs, run code, or take actions, you see exactly what it is doing and what results it gets.
- Artifacts: Generated outputs like charts, tables, and code that the assistant creates during investigation.
Every conversation is saved as a thread that you can return to later, share with colleagues, or fork to explore different hypotheses.
What you can ask
The assistant can help with a wide range of debugging and investigation tasks.
Debugging
Ask about errors and failures in your system.
Why are users seeing 500 errors on the checkout page?
What is causing the login service to timeout?
Show me the stack traces for the NullPointerException we saw this morning.Investigation
Correlate events and understand what changed.
What changed in the last hour that could explain this latency spike?
Which deployment caused the memory usage to increase?
Are these errors related to the database migration we ran yesterday?Monitoring
Get visibility into trends and patterns.
Show me error trends for the payment service over the last week.
What is the p99 latency for API requests today?
Are there any anomalies in the logs from the last 24 hours?Code fixes
Get help understanding and fixing issues in your code.
Help me fix this null pointer exception based on the stack trace.
What would cause this race condition in the order processing code?
Review this error handler and suggest improvements.How the assistant thinks
The assistant does not guess. It investigates.
When you ask a question, the assistant:
- Plans: Determines what information it needs to answer your question.
- Gathers evidence: Uses tools to search logs, run queries, and fetch data.
- Analyzes: Correlates findings, identifies patterns, and interprets results.
- Explains: Reports what it found with specific evidence and reasoning.
You see each step as it happens. The assistant shows which tools it calls, what queries it runs, and how it interprets the results. This transparency lets you verify its reasoning and guide the investigation when needed.
The assistant will not make claims without evidence. If it cannot find relevant data, it tells you what it searched and suggests alternative approaches.
Key capabilities
Explore the assistant's capabilities in depth:
Tools
The assistant can search logs, create artifacts, manage alerts, and connect to external services through MCP.
Delegation
Complex investigations can spawn subagents that work in parallel, and background investigations can run autonomously.
Sandbox
When the assistant needs to run code, it uses a secure sandbox environment that you can customize with init scripts and environment variables.
Effective Prompting
Learn techniques for writing prompts that get better results from the assistant.