Skip to main content
Our proprietary engine consolidates all your connected apps and uploaded data into a unified system that agents can interact with through three core tools: manifest, search, and operate. These tools provide intelligent routing, parallel processing, and approval-based workflows across all your integrations. You can also allow agents to discover and invoke specific tools from an integration (say, JIRA_LIST_TICKETS) without going through the engine. See Passthrough Tools. Our Engine For Reference Only You don’t have to memorize how these tools work or what parameters they require. Your agent will handle all the technical details automatically. This documentation is here for reference when you need it. Searches apps, files, and webpages, then returns ranked results with scores and metadata. Our engine intelligently routes queries to the most relevant integrations based on the search content. Integrations are searched in parallel with early termination when sufficient results are found.
Source types:
  • Apps: Connected integrations (Slack, GitHub, etc.) in the workspaces the Agent Key can access.
  • Files: Documents uploaded to folders the Agent Key’s workspaces can access.
  • Webpages: Documentation pages crawled in the folders the Agent Key’s workspaces can access.
Input
ParameterTypeDefaultDescription
querystring—Natural language search query
source_typesstring"all"”all” or comma-separated list like “app,file,webpage”
integration_idsstring""Leave empty to search all, or specify UUIDs for specific integrations
limitnumber10Maximum results per source type
search example

​
operate

Plans and executes multi-step tasks through an approval-based workflow. This enables complex operations that require multiple actions like “yo, look at my open tickets, compare them to the sprint board, and sync the status.”
How it works:
  1. Compilation: Your agent tells Commons it’s intent. Commons compiles a plan based on available integrations and knowledge and returns it to your agent.
  2. Reviewal: Your agent recieves and reviews a structured breakdown of phases and actions that will be executed. Your agent has final approval authority.
  3. Execution: Your agent resumes with approved=True to execute the next phase. The tool executes one phase at a time and waits for approval before continuing.
  4. Completion: Steps repeat until all phases are executed or your agent aborts the operation.
Mid-execution control:
  • Your agent can modify future phases by passing phase_modifications to add, remove, or change planned actions.
  • When Commons pauses for additional information (e.g. “I found two Claire’s — who do you want to Slack?”), your agent provides it via user_input and execution continues.
  • Your agent can request an execution log with get_full_details=True to see all completed actions and their results.
Input
ParameterTypeDefaultDescription
descriptionstring—Natural language description of what to accomplish (initial call only)
job_idstring—Job ID for resuming execution (continuation calls)
approvedbooleanFalseSet to True to execute the next phase
user_inputobject—Data to provide when execution needs user input
phase_modificationsobject—Changes to add, remove, or update future phases
get_full_detailsbooleanFalseReturn complete execution log
operate example

​
manifest

Agents call manifest to get a complete inventory of everything accessible with your Agent Key. This returns all workspaces, connected integrations, uploaded files with summaries and metadata, and passthrough tools information if enabled.
InputNo parameters required.The manifest tool is simply a utility function for your agent to get a lay of the land. The example call below was done in our own Sandbox. The Sandbox agent confirms it has access to a sample workspace. Visit the Sandbox page to learn more.manifest example