> ## Documentation Index
> Fetch the complete documentation index at: https://docs.erstan.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Tools overview
> What Erstan tools are, where they run, and how they are grouped by source.
Tools are deterministic capabilities an agent can call while it works. A tool can read data, create an Erstan artifact, update a task, send a message, run SuiteQL, call a connector API, or invoke a provider built-in such as web search.
Tools are different from skills and connectors:
| Concept | What it is | Example |
| ------------- | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------ |
| **Tool** | One operation the runtime can call. | `er_task_create`, `er_document_read`, `ns_runCustomSuiteQL`, `web_search`. |
| **Connector** | The source and authorization behind a set of tools. | Erstan, NetSuite MCP, Slack, OpenAI, Anthropic, or an MCP server. |
| **Skill** | Reusable instructions and optional package files that guide how an agent should work. | A NetSuite close skill that tells the agent which reports, tools, and checks to use. |
## Where tools run
Tools are available across Erstan, but the control surface changes by context.
| Surface | How tools are used |
| ----------------- | --------------------------------------------------------------------------------------------------------------------------- |
| **Chat** | Erstan chooses eligible tools based on the message, selected connections, available models, and write approvals. |
| **Agent Builder** | Builders pin tools on AI nodes, add By Connector action steps, or use AI Plan & Execute for a bounded tool-and-answer step. |
| **Tasks** | Task-backed runs use Erstan task, project, document, file, and comment tools to keep work visible. |
| **Connectors** | Owners and admins review connector tools, connection health, and per-tool write policy from each connector's **Tools** tab. |
## Tool sources
Erstan groups tools by source so builders can tell what owns the operation and which authorization model applies.
First-party tools built into Erstan for tasks, projects, documents, files, workbooks, teams, skills, agent runs, Agent Builder, and SuiteQL helpers.
Tools exposed by connected systems or providers, including NetSuite MCP, Slack, OpenAI, Anthropic, business apps, and custom MCP servers.
## Read, write, notify, and attach behavior
Every tool has an operational shape. The shape is not only documentation: it informs approvals, connector policy, and how you design an agent flow.
| Behavior | Meaning | Examples |
| ---------- | ------------------------------------------------------------------------- | ------------------------------------------------------------------ |
| **Read** | Fetches or analyzes data without changing source records. | Search tasks, read a document, load a file, run read-only SuiteQL. |
| **Write** | Creates, updates, appends, archives, assigns, or otherwise changes state. | Create a task, update a document, create a NetSuite record. |
| **Notify** | Sends a message or email outside the current run. | Send an Erstan email or post to Slack. |
| **Attach** | Links an artifact, file, document, URL, or reference to existing work. | Attach a task reference. |
Write-capable connector tools should stay on **Require approval** unless your team has deliberately tested and accepted the automation risk. NetSuite writes are especially sensitive because they can change ERP records.
## Tool access in Agent Builder
AI steps can be configured with different levels of tool access:
| Policy | Use it when |
| ---------------- | -------------------------------------------------------------------------------------------------------------------------------------------- |
| **No tools** | The step should only transform or respond using existing context. Use this for final summaries, review text, and formatting-only steps. |
| **Pinned tools** | The step must use a specific, reviewed tool set. Use this for production flows, deterministic lookups, ERP work, and any write-capable path. |
| **Auto** | The step can discover low-risk tools dynamically. Use this only for exploratory or general assistance. |
For production agents, pin the exact tools a step needs and describe when to use them in the step instructions.
## Managing access
Owners and admins control connector availability and per-tool write policy from **Connectors**. Builders control which tools a specific agent step can use. End users approve write actions when a policy or agent step requires a human checkpoint.
For Agent Builder guidance, see [Choosing AI steps](/agents/ai-steps) and [Giving agents tools, skills & connections](/agents/tools-skills-connections).
## Next steps
Review the first-party Erstan tool catalog.
Review connector-backed and provider-backed tools.
Decide when an agent should think, respond, call tools, branch, loop, or pause.
Connect systems and manage tool write policy.