Skip to content

Tools Reference

This reference documents the tools that agents can use, organised by category. Tools are the actions agents take during conversations — searching knowledge, sending messages, reading files, running code, and working with integrations. You will see these tool names in the activity log when an agent works on your request.

Search your organisation’s knowledge bases using keyword and semantic search. Returns matching entries ranked by relevance.

Key parameters: query (search text), index (optional — target a specific knowledge base), limit (max results)

Write a new entry to a knowledge base or update an existing one. Use this to store information the agent discovers during conversations.

Key parameters: index (target knowledge base), title, content, keywords (list of related terms for discoverability)

Work with Collections — typed, faceted datasets your organisation defines (for example, a customer register or an asset list). Supports defining, querying, and ingesting records.

Pin durable facts and open tasks to the agent’s working memory so they persist across turns without re-searching.

Let the agent manage its own working memory — keeping load-bearing content, dropping stale facts, and clearing its working set when the topic changes.

Analyse and store communication style preferences so the agent can match how you like to be spoken to.


Send a message to another agent. This is how agents collaborate — delegating tasks, asking questions, and sharing information.

Key parameters: agent (recipient agent name or ID), message (the content to send)

Ask another agent to perform a task and receive the result inline. Available to every agent, not just companions — any agent can consult a specialist mid-conversation. Self-delegation is blocked.

Key parameters: agent (target agent), task (description of what needs to be done)

Discover which agents in the organisation can help with a given topic. Agents use this before delegating so requests land with the right specialist.

Proactively send a direct message to a person, or post into a group or channel the agent belongs to.

Create a new group or channel in team chat, ready for the agent to post into.

Search your organisation’s team chat history. Companion agents search on their user’s behalf; shared agents have a narrower scope.


Pause and ask the user to make a decision — yes/no or a multiple-choice pick — before continuing.

Show the user an interactive panel (a form, preview, or editable surface). The user’s edits come back to the agent as a submission.

Offer the user clickable follow-up actions at the end of a reply.

Ask the user to approve access to an external web domain the agent is not yet allowed to call, instead of silently failing.


Create a new specialist agent on the fly. Only companion agents have this tool, since creating agents is a privileged action performed on a user’s behalf. The companion sets up the agent with an appropriate persona and settings based on your description.

Key parameters: description (what the new agent should do), name (optional agent name)


See Schedules & Tasks for how scheduled tasks work end to end.

Create a deferred task — a recurring schedule (cron-based) or a one-off task that fires once at a specific time. Supports verbatim reminders, natural-language instructions the agent carries out, data pipelines, and sandboxed scripts.

Key parameters: name, action (what to do), cron_expression (recurring) or fire_at (one-off), task_type (reminder, message, script, or pipe), notify_channel (where to deliver output)

Modify an existing scheduled task — reschedule it, rename it, pause or resume it, or edit its instructions.

List all scheduled tasks for the current agent, including timing and next fire time.

Remove a scheduled task so it no longer runs.

Key parameters: schedule_id (the schedule to remove)

Have the agent check back in after a set number of minutes — a lightweight “poke me later” for follow-ups within a conversation.


These tools let an agent wait for something to happen without holding a conversation open.

Poll an HTTP endpoint on an interval until a condition is met, then resume.

Kick off a long-running tool operation in the background and pick up the result when it finishes.

Subscribe to a platform event and resume when it fires.

List or cancel the agent’s active polls and event waits.


Mark a step in a multi-step workflow as complete, optionally providing output data for the next step.

Key parameters: step_id, output (result data)

Mark a workflow step as failed with a reason, so the workflow can handle the failure.

Approve or reject a workflow step that is waiting on review. Available to the agent assigned as the step’s reviewer.

Run a background task with its own retries and checkpointing, separate from the current conversation.

A powerful tool with 40+ actions for managing your organisation. Actions include managing users, agents, teams, knowledge bases, integrations, and settings. This is also where agents call create_workflow, update_workflow, and validate_workflow when authoring workflows in chat.

Key parameters: action (the specific management action), plus action-specific parameters


The agent’s persistent file workspace — read, write, and organise working files that survive across conversations.

Read the contents of an uploaded file. Supports documents, spreadsheets, and text files.

Key parameters: file_id

Manage files — list, retrieve, and share files associated with conversations or the organisation.

Create and edit office documents — Word, Excel, PowerPoint, and PDF generation and assembly.

Render content (Markdown or SVG) into a PDF file.

Render an SVG into an image file.

Split a large PDF into parts the agent can read, or merge an ordered list of workspace PDFs into a single file.


Every agent can write and run code — these tools are always available.

Run code in a persistent Python or JavaScript interpreter to perform calculations, data transformations, or custom logic during a conversation.

Key parameters: language (python or javascript), code (the code to run)

Write a script that orchestrates many tool calls in one pass — useful for bulk operations that would otherwise take many conversational turns. Integration tools are callable from inside the script.

Run shell commands and work with files in the agent’s sandbox environment.

Work with git repositories in the sandbox and push changes to Sprigr-hosted repositories.

Delegate a whole multi-step assignment (code, documents, data pipelines) to an autonomous coding agent that runs headless in the sandbox with shell and workspace access. See Deep Tasks for how these runs work, their budgets, and the write-access approval flow.

Edit your organisation’s Sprigr-hosted WordPress site — the agent resolves the site automatically, so a request like “change my store’s homepage” just works.


Browse websites and extract content. Supports navigating pages, clicking, filling forms, searching, and pulling information from public web pages.

Key parameters: action (for example navigate, find, click, fill, search, parse), url

Make a direct HTTP request to an external API. Useful for services that do not have a dedicated integration. Outbound domains are restricted; the agent can ask you to approve new ones via request_outbound_access.

Key parameters: method (GET, POST, PUT, DELETE), url, headers, body


list_email_templates / apply_email_template

Section titled “list_email_templates / apply_email_template”

List your organisation’s saved email templates and apply one (including the configured signature) when composing.

Look up a contact in your connected helpdesk or CRM to add context about who the agent is talking to. Returns a “no match” message when nothing is connected.


File a report with the Sprigr platform team — bug reports, feedback, or issues the agent hits while working. Available to every agent.

Key parameters: title, content

Manage conversation threads — list, read, and organise the agent’s inbox.

Perform administrative platform operations. Only useful to agents with elevated access.


When integrations are connected to your organisation or a specific agent, additional tools become available. Below is a summary of tools provided by each built-in integration.

ToolDescription
gmail_searchSearch emails by query, sender, date, or label
gmail_readRead the full content of a specific email
gmail_sendCompose and send a new email
gmail_deleteDelete an email
gmail_download_attachmentDownload an attachment from an email
gmail_contacts_searchSearch your Google contacts
ToolDescription
calendar_list_eventsList calendar events for a date range
calendar_create_eventCreate a new calendar event
calendar_update_eventUpdate an existing event
calendar_cancel_eventCancel an event
calendar_check_availabilityCheck free/busy availability
ToolDescription
drive_list_filesList files in a folder
drive_search_filesSearch for files and folders
drive_read_fileRead the content of a document or file
drive_upload_fileUpload a file to Drive
drive_create_folderCreate a folder
drive_move_fileMove a file between folders
drive_delete_fileDelete a file
ToolDescription
sheets_list_spreadsheetsList available spreadsheets
sheets_read_rangeRead a cell range
sheets_append_rowAppend a row to a sheet
sheets_update_rangeUpdate a cell range
sheets_create_spreadsheetCreate a new spreadsheet
sheets_searchSearch within spreadsheets
ToolDescription
docs_readRead a document
docs_create_from_templateCreate a document from a template
docs_export_pdfExport a document as PDF
ToolDescription
outlook_mail_searchSearch Outlook emails
outlook_mail_listList emails in a folder
outlook_mail_readRead a specific email
outlook_mail_sendSend a new email
outlook_mail_replyReply to an email thread
outlook_mail_forwardForward an email
outlook_mail_moveMove an email between folders
outlook_mail_deleteDelete an email
outlook_mail_download_attachmentDownload an attachment
outlook_contacts_searchSearch your Outlook contacts
ToolDescription
outlook_calendar_list_eventsList calendar events for a date range
outlook_calendar_create_eventCreate a new calendar event
outlook_calendar_update_eventUpdate an existing event
outlook_calendar_cancel_eventCancel an event
outlook_calendar_check_availabilityCheck free/busy availability
outlook_calendar_find_meeting_timesSuggest meeting times that work for attendees
ToolDescription
onedrive_list_filesList files and folders
onedrive_read_fileRead a file
onedrive_upload_fileUpload a file
onedrive_create_folderCreate a folder
onedrive_copy_itemCopy a file or folder
onedrive_delete_itemDelete a file or folder
onedrive_list_shared_with_meList items shared with you
onedrive_read_shared_fileRead a shared file
onedrive_excel_read_rangeRead a range from an Excel workbook
onedrive_excel_update_rangeUpdate a range in an Excel workbook
onedrive_excel_add_worksheetAdd a worksheet to a workbook
onedrive_excel_searchSearch within Excel workbooks
ToolDescription
teams_send_channel_messagePost a message to a Teams channel
teams_send_chat_messageSend a message in a Teams chat

GitHub is a single github tool that takes an action parameter:

CategoryActions
Repositorieslist_repos, get_repo, read_file, create_or_update_file, search_code
Branches and commitscreate_branch, list_branches, list_commits
Issueslist_issues, create_issue, update_issue, add_issue_comment
Pull requestslist_pull_requests, get_pull_request, create_pull_request, add_pull_request_comment

Gorgias is a single gorgias tool that takes an action parameter:

CategoryActions
Ticketslist_tickets, get_ticket, create_ticket, update_ticket, reply_to_ticket
Customerssearch_customers, get_customer
Tagslist_tags

Replies default to internal notes for safety; the agent must explicitly choose a customer-facing reply.

Amazon Web Services is a single aws tool with around 30 actions covering common services:

CategoryActions
EC2List, describe, start, stop, and reboot instances; read console output
LambdaList and inspect functions, invoke, read logs and metrics, update config
S3List buckets, list objects, get and put objects
CloudWatchList alarms, get metrics, query logs, create alarms
Cost ExplorerGet costs, forecasts, and savings recommendations
SSMRun commands on managed instances and read output
IAM / STSList users and roles, account summary, verify identity

Destructive actions (stop, reboot, invoke, put, run command) require user confirmation, and the integration can be restricted to read-only access, specific regions, and specific services.

Google Cloud Platform is a single gcp tool with around 40 actions covering common services:

CategoryActions
Compute EngineList, describe, start, stop, and reset instances
Cloud FunctionsList, describe, invoke, read logs, update config
Cloud StorageList buckets, list objects, get, upload, and delete objects
Cloud RunList and describe services, list revisions, read logs
BigQueryList datasets and tables, run queries, get job results
Monitoring and LoggingGet metrics, list alert policies and uptime checks, query logs
BillingGet billing info, budgets, and costs
IAM / GKEList service accounts and roles, get policy, list clusters and workloads

Like AWS, the GCP integration can be restricted to read-only access.

Beyond the built-in integrations, agents gain tools from two other sources:

  • Marketplace apps — Apps installed from the Sprigr marketplace add their own tools. For example, the Atlassian app adds jira_search, jira_create_issue, jira_update_issue, jira_add_comment, jira_transition_issue, confluence_search, confluence_get_page, confluence_create_page, and more. Slack and Shopify are also delivered as marketplace apps.
  • MCP servers — Connecting a Model Context Protocol server registers every tool the server exposes, so any MCP-compatible service becomes available to your agents.
  • Creating Agents — Create an agent and assign it the tools it needs.
  • Agent Settings — Configure integrations and tool access for each agent.
  • Companion Agents — Learn about the companion-only create_specialist tool.
  • Knowledge Bases — Understand how sprigr_search and sprigr_write interact with your knowledge bases.