Agent tools
The tools the Sazabi agent uses during an investigation — log search, web search, your code, issues, and integrations — and how to configure which are available.
The Sazabi agent investigates by using tools, and it shows you every tool call it makes. This page explains the main tools and how to control which ones are available in your project. You configure tool availability under Settings > Tools.
Log search
Searching your logs is the agent's core tool and its default. When your project has log storage, the agent can query your logs to answer questions — filtering by service, severity, time range, and content — and it shows the query it ran so you can verify the result. This is what lets you ask "what errors is checkout throwing?" and get an answer grounded in real records.
Web search
The agent can search the public web to bring in outside context — for example, what a specific error message from a dependency means.
- Web search is controlled per project under Settings > Tools. It is off unless you enable it.
- You can restrict which domains the agent may search with allow and deny rules. Organization rules apply everywhere, and a project can override them.
Recommended: enable web search when your investigations benefit from looking up error signatures or vendor documentation. Use domain rules when you want to keep the agent's web access to a known set of sources. Trade-offs: web search adds outside context but reaches beyond your own data. Do not enable it if your policy requires investigations to stay grounded only in your telemetry and code.
Your code
When you connect GitHub or Bitbucket under Settings > Code, the agent can read the repositories you link — so it can trace an error to the file and line responsible and reference your actual code in its answer. This is repository access you grant, not a public code index. See Instrument, GitHub integration, and Bitbucket integration for connecting code.
Issues
The agent can work with your issues during an investigation: search existing issues, open a new one when it finds a problem worth tracking, and update an issue's status. This is how an investigation turns into something your team can track and act on.
MCP connectors
Through MCP connectors, the agent can reach the external tools your team uses — for example creating a ticket in your issue tracker or reading from a connected service. Connectors are configured at the organization level under Settings > MCP connectors, and the tools they expose become available to the agent. Giving the agent tools this way is the opposite direction from using Sazabi from your coding agent, which connects an outside agent to Sazabi.
Sandbox & CLIs
The agent works in a per-project sandbox where it can run shell commands and use command-line tools. That includes Bash for ad-hoc commands, the command-line tools you connect from the sandbox CLI catalog or your own custom CLIs (database clients, cloud and platform CLIs), and the Sazabi CLI itself for querying your own logs and project data. Connect the CLIs a project should have under Settings > Sandbox CLIs; the agent uses whatever is connected.
Memory
The agent can read and write project memory — durable notes and runbooks scoped to the project — so what it learns in one investigation is available in the next. It reads relevant memory when it starts and can record findings back.
Configuring available tools
Go to Settings > Tools to control which tools the agent may use in a project — for example turning web search on or off and setting its domain rules. Changes apply to new investigations in that project.