6 min read

Tools

Tools are the actions an agent can take during a run — read a record, send an email, search the web, call an integration, and more. You choose which tools each agent has on its Tools tab in agent settings.

Enabling tools for an agent

Open an agent's settings and go to the Tools tab. Tools are grouped by category — toggle on only what that agent needs. An agent cannot call a tool you haven't enabled.

A practical starting point: Read records first, then add write, email, or other tools as you validate behaviour. Connected integrations appear as separate toggles once they're set up at the workspace level.

Tool groups

Data

Read records, write records (create/update/delete), and manage data connections (inspect syncs, change schedules, trigger resyncs, prompt users to connect).

Productivity

Manage tasks, schedule/list/cancel reminders, read and run skills (agentskills.io files), and search across the agent's other threads.

Memory & Knowledge

Private agent memory (remember/recall/forget) plus the shared knowledge graph (search entities, record events, link facts).

Communication

Check and send email, read/send Slack messages, send WhatsApp messages, verify channel identity, and call other agents.

Browsing

Search the open web, browse within a domain, and computer use (remote browser with live view in the thread).

Research

Semantic web search, deep topic research with citations, quick sourced answers, and structured dataset extraction — billed via LLM credits.

Network & code

Send webhooks, fetch URLs, run JavaScript in a sandbox, and run Python (when E2B is configured).

Files

Read spreadsheets and CSV (XLSX, XLS, TSV, ODS), extract text from Word (.docx) and PDF, inspect images and documents, and share generated files via 7-day signed download links — all always on, no grant required.

Multimedia

Generate images inline in the thread (billed via LLM credits).

Advanced

Modify schema, update app pages and navigation, create or update other agents, and author skills. Enable only when an agent genuinely needs these capabilities.

Reading and sharing files

Several file tools are always on for every agent — you don't toggle them on the Tools tab. They work on URLs the agent already has access to: thread attachments, record file fields, and signed links. Parsing runs locally with no LLM call.

Read spreadsheets and CSV

The read_spreadsheet tool parses CSV, XLSX, XLS, TSV, and ODS into structured rows so the agent can filter, total, or turn a sheet into records — without burning credits on a vision call.

Extract Word text

The extract_docx_text tool pulls plain text from modern .docx files (not legacy binary .doc). Use inspect_document instead when layout, tables, or images matter.

Share files with a download link

The share_file tool saves content you generate (a CSV export, report, JSON dump) and returns a signed download link valid for 7 days. Pass the link in chat or to send_email as an attachment.

The same share capability is available outside agent threads: skill code can call stacker.share_file (or shareFile in JavaScript), and automations use stacker.shareFile({ ... }) — including runs triggered by record changes or buttons that have no agent thread. See Automations for the SDK shape.

Automations

Alpha — agent automations are in early access and not on by default. To try Run automations or Modify workflows, schedule a call or email support@stackerhq.com to chat about it.

When enabled for your workspace, agents can trigger app automations by slug and help author or update workflow code — start a thread, describe what you want in plain language, and the agent builds the automation the same way you'd ask it to update a record or the portal builder to change a page. Automations run JavaScript or Python in a sandbox on record changes, button clicks, agent calls, and other triggers.

See Automations for how triggers, callable workflows, and the Stacker SDK work once you have access.

Requesting more tools mid-chat

With Request more tools enabled on the agent, it can ask the user to approve an additional tool grant — once, for the thread, or permanently. The thread shows exactly which tool is being requested before anything changes. See Starting threads for examples.

Data, connections & integrations Back to overview