Running a Chat Assistant

A flow that starts at Robomotion.ChatAssistant.ChatIn does nothing on its own: it waits for a message from the Chat Assistant page. robomotion run would start it and watch it wait until --timeout. So a Chat Assistant flow is tested the way its users meet it: as an Agent in the Workspace, running on its own robot, talked to in the real Chat Assistant page.

By hand that is the Designer (version, publish), Admin Console > Agents (create, one instance), the Desktop App (connect the agent's robot, Play) and the /agents page (Open), then typing. robomotion agent does all of it from the terminal:

robomotion validate .                        # 1. pspec-clean first, always
robomotion agent push                        # 2. this folder's flow → "<name> (dev)" in the workspace
robomotion agent create --mode guided        # 3. the agent with one instance (or --mode conversational); publishes a version itself if there is none
robomotion agent start                       # 4. connect the agent's own robot on this machine, then Play
robomotion agent logs -n 20                  #    wait for "Started running <name> [...]" before chatting
robomotion agent chat "…" --expect "…"       # 5. the real page, one step at a time, every reply printed

In a folder that is a git checkout of its cloud flow, agent push refuses (saving there is git). Step 2 is then git add -A && git commit -m "…" && git push; steps 3 to 5 stay the same. Everywhere else, agent push is step 2. After an edit, see Step 4.

What you need: the robomotion CLI on PATH. It brings robomotion-deskbot (the robot) and robomotion-browser-mcp (the browser) with it. Plus a login made with robomotion auth login in the flow folder (a session, not an API key: the chat page signs in with it). Nothing else: no Designer, no Admin Console, no Desktop App, no Playwright, no curl.