deepseek messages transport failed
Transport failed means the request to the model provider never completed — the session is intact, the network path or the provider config is not.
The message appears in the terminal that runs the profile, not in the browser tab.
The error, verbatim
It arrives as a failed request on the model call, and the session keeps the events that led up to it.
What usually causes it, most likely first
A proxy or VPN that closes long-lived streaming responses. DSH streams model output; an intermediary with a 30-second idle timeout kills the connection mid-answer.
An expired or revoked provider key. The request leaves, the provider rejects it before any token is produced, and the client reports a transport failure rather than an auth error.
A model id that the profile does not actually configure. Nothing listens on that name, so no response arrives.
One command to see which provider and model the profile uses
Read the composed tree instead of guessing which layer set the model.
dsh --profile web --dump-configIf that does not settle it
Try the same profile from a network without the proxy. If it works there, the proxy is the cause and the fix is a stream-friendly setting, not a DSH setting.
If it fails everywhere, check the key: re-run the provider login flow. Do not edit credential files by hand.
Common questions
Does this corrupt the session?
No. The failed request is an event like any other; the log keeps what happened before it.
Why not just a timeout message?
Because the failure is reported at the transport layer, before the provider produces a first token — there is no partial answer to attribute a timeout to.