Troubleshooting#

Common problems when launching and using Agent Master, grouped by where they occur.

Launch#

Dash / app dependencies missing.

Install the application extra and relaunch:

pip install "pycsamt[app]"

Port 8765 already in use.

Another process (or a previous Agent Master) holds the port. Launch on a different one:

pycsamt-agent --port 8770

The browser does not open.

On headless or remote hosts there may be no browser to open. Launch with --no-browser and browse to the printed URL (http://127.0.0.1:8765 by default) yourself. From another machine on a trusted network, use --host 0.0.0.0 and the server’s address.

The terminal exits / nothing serves.

The terminal running pycsamt-agent hosts the server — keep it open. Closing it stops the app.

Model Provider#

A request needs a model / “no API key”.: The language-driven parts need a provider and key. Open Settings, choose a provider, paste the API key, set the model and the active provider, then Save API keys (see LLM Configuration). Resend the request.
Provider connection errors.: Check the key, the selected model id, and network access to the provider. A proxy or firewall may block the provider’s endpoint. Switching the active provider to one you have a working key for is a quick workaround.

Data And Workflows#

Nothing runs / the assistant asks for data.: Load an EDI survey first with Load EDI — every workflow operates on a loaded survey (see Welcome Screen And Chat).
A workflow keeps asking which lines to process.: That is expected when several lines are loaded: choose lines in the Select line(s) to process panel and Run selected, or Run all lines (see Workflows And Agents).
A workflow fails partway.: Read the steps trace in the response for the failing step, adjust the parameters in the in-chat form (method, thresholds, period limits), and run again.
Generated code will not import.: Agent Master validates generated scripts before showing them, so imports should resolve. If a script still fails, confirm the pyCSAMT version in the environment you run it in matches the one that generated it, and that the data path in the script header points at your survey.

Session And Browser#

A new tab starts empty.: Sessions are per browser context. Use Save and the Session tab to persist and restore a conversation rather than relying on a second tab.
Stale state after an update.: Start a New Chat, or reload the page; re-load the survey if the loaded badge looks wrong.

When You Need More Detail#

Relaunch with dev tools for tracebacks in the terminal and browser:

pycsamt-agent --debug

Use --debug only for local diagnosis; run without it for normal use.