In an AI-assisted delivery engagement, people and AI agents both do real work: shaping requirements, designing, testing, and changing the system. The hard part is keeping everyone on the same page about what the system does and what was agreed. Catalio gives people and agents one shared specification to work from, from gathering requirements to proving that what ships behaves as intended.
One shared specification for people and agents
Catalio keeps your living specification of each application: capabilities, processes, features, business rules, use cases, and requirements. Your team reviews and confirms it, and agents read the same catalog when they answer you. Nobody works from a private copy.
Agents you work with
Every organization has the same five agents, each with its own area: Product Manager, Architect, QA Analyst, Security Analyst, and DevOps. Every agent can read your catalog. Each one creates or changes catalog records only in the areas its role is allowed to write. In the command center chat header, click the agent’s mark or name to open its profile. On Home, the mark and name at the top switch agents; open the profile from Agent profile in the switcher. On a phone, tap the status line under the conversation title. See AI Chat.
Agents hand work to each other
When you ask an agent for something outside its area, it can hand that part to the right agent. You don’t assign the work yourself; the agent starts the handoff, and you stay in the loop. A handoff shows in your conversation as a card with both agents and its status: Waiting, Answered, Not answered, or Stopped. Click the card to read what was asked. The Stop button ends handoffs in that conversation. Handoffs stay within your own conversations. See AI Chat.
Give agents the right context
In a conversation with an agent, type @ in the message box to add a record to the conversation’s context: a requirement, feature, application, component, or initiative you can see. The agent keeps it in mind for the rest of the conversation. (You @mention records, not agents.)
Each agent also has a Knowledge tab. Org holds notes and docs your organization keeps for that agent, which only an organization admin can change. Mine holds your own notes and docs, which only you and the agent see. See Page Context and Agent Knowledge.
Your tools, with people in control
An organization admin adds MCP servers in Settings → Connections, and each member signs in with their own account. Agents can then use those tools in chat. Before an agent deletes a record, removes a link, merges records, rewrites PRD sections, or uses a server tool that isn’t marked read-only or non-destructive, it asks: Run this change? Click Approve to go ahead or Deny to skip it. Creates and updates run without asking. See MCP Connections and What the AI Can Do.
Work from your IDE
Developers can add Catalio as a remote MCP server in Claude Code, Claude Desktop, or any MCP-compatible client, and work with the same specification without leaving their editor. See MCP Authentication and MCP Technical Workflow.
Gather requirements
Most engagements don’t start from a blank page. With Extract, Catalio reads your code, documents, and conversations and drafts the application’s specification. Nothing is final until someone on your team reviews it. See Extract overview.
Verify behavior
Link test cases to requirements and record each run as a test result: Passed, Failed, Skipped, or Blocked. Over time those results show which requirements have been verified. When you are replacing a system, Gap Analysis compares what the current system does with what the target system needs to do. See Test Cases.
