The recording pairs a fresh profile through stdin, starts a Luna thread, waits for its answer, continues the conversation, and lists the selected project and thread.

Run a conversation from your terminal

t3c sends prompts to a stock T3 Code server and prints the assistant’s final answer. Use it in a terminal or a harness that supplies a prompt on stdin. Node >=22.16 is required. The target is T3 Code 0.1.9, protocol 2.

Install with npm or Homebrew:

npm install -g @wyrd-company/t3code-cli
# or
brew install wyrd-company/tools/t3c

On the server, use upstream t3 pair or t3 auth pairing create to obtain a pairing link. Pair it through t3c auth pair '<pairing-link>'. For an issued session token, use stdin so the token stays out of the argument list:

t3c auth pair --server https://server.example --token-stdin --name example

Supply the token through stdin from your existing secret source. The command validates it before saving a profile. See auth and profiles.

t3c version
t3c run 'Reply with one short greeting.' --model gpt-6-luna
printf 'List three basic colors.\n' | t3c run --model gpt-6-luna
t3c thread list
t3c thread read <thread-id> --last
t3c run --thread <thread-id> 'Put the colors in alphabetical order.'

run waits without a default run deadline. Add --timeout 5m when needed. Project paths are paths on the server. --project-path /workspace/project selects that project; omission uses the server Scratch project. Provider and model selection use project defaults, then server defaults. Ambiguous selection asks for an explicit choice through a usage error.

Shared servers

t3c uses the direct route: pair its profile with the server’s own address. If people share the server through t3c-proxy, do not pair t3c with a person’s hostname. A message that t3c sends there ends with that person’s byline.

Use command behavior, output and exit codes, the Dagu harness, the library guides, or the library API for details. t3c --help and each command’s --help list supported options.