The full set of capabilities agents can invoke mid-workflow to complete a task end-to-end without human intervention.
Book a demoActions an agent can take during or beyond conversation - looking something up, reading a document, browsing a page, writing to a system, or even sending a message. Rather than handing off to a human when a task requires more than talking, agents invoke tools directly within the workflow exactly when needed.
Agents aren't limited to responding in the channel where a conversation started. Mid-workflow, an agent can send an SMS confirmation, fire an email summary, post a message in Slack or Teams, without breaking the conversation flow.
Agents can receive and process inbound messages across channels as part of a workflow - an SMS, an email, a form submission and more, and continue execution based on what comes back. This enables multi-turn, multi-channel interactions spanning several sessions.
Agents can receive and process file uploads of any type including documents, images, audio, and video sent by a user or passed in from an external system, and act on their contents within the workflow.
/ 01
Agents invoke OCR mid-workflow to read printed or handwritten content from images, PDFs, and scanned documents. Extracted text is parsed into structured fields and passed directly into the agent's reasoning or written to a downstream system - for invoices, bills of landing, identity documents, forms, and any document that arrives as an image.
/ 02
Not every system an enterprise operation depends on has an API. Browser agents allow a workflow to navigate a web interface, locate information, fill fields, and extract data the same way a human operator would - so legacy tools, supplier portals, and carrier websites can be incorporated into automated workflows without a custom integration.
/ 03
Agents retrieve content from knowledge bases at the point in a conversation where it's needed, matching context to the most relevant content available. Knowledge bases support SOPs, documentation, policies, contracts, and training materials - indexed so agents retrieve precisely rather than broadly, and updated without redeploying the agent.
/ 04
Agents can analyze the content of images and documents - not just extract text, but interpret what's in them, across any channel. A photograph of a damaged shipment sent over WhatsApp, a scanned certificate emailed in, a product image uploaded via chat - agents reason about visual content wherever it arrives and act on it as part of a workflow decision.
/ 05
Agents can parse API responses, transform data structures, apply business rules, and compute values mid-workflow using code blocks with full programmatic control at any point in execution without leaving the workflow.
/ 06
When a workflow requires a capability that lives in another agent, such as an existing enterprise investment, a specialized model, or a third-party system, the agent can delegate to it via MCP, pass context, wait for a response, and continue seamlessly. Existing agent infrastructure plugs directly into the orchestration without being rebuilt.
