help

Tritium is our coding agent for developers, vibe coders, and agents.
Pseudocode input, Tritium converts to Python / C / Rust.
Back to Tritium

You write Pseudocode. Mark bodies Tritium should fill with a Tritium token (@tritium). Tritium converts to Python / C / Rust via metadata.lang / --lang. Details: guide.

Primary path: write Pseudocode + @tritium and convert with the CLI — same as demos.

pip install ultcan-tritium --index-url https://ultcan.com/simple/

PSEUDO='def shortest(graph, source):
    """@tritium
    fn shortest(graph, source):
      return distances from source on the weighted graph
    """
    ...'
tritium -c "$PSEUDO" --lang python

Same Pseudocode over HTTPS (POST /v1/chat/completions, metadata.lang = python | c | rust):

BASE='https://ultcan.com/tritium/api'
curl -sS "$BASE/v1/chat/completions" \
  -H 'Content-Type: application/json' \
  -d '{"model":"tritium","metadata":{"lang":"python"},"messages":[{"role":"user","content":"def shortest(graph, source):\n    \"\"\"@tritium\n    fn shortest(graph, source):\n      return distances from source on the weighted graph\n    \"\"\"\n    ..."}]}'

HTTP	meaning
200, completion_tokens=0	Converted code in the assistant message
422	Lower Pseudocode until HIT — guide
429	Rate limited — wait retry_after_s

Optional chat commands (same endpoint, message body is the command word):

message	what it does
help	this card (from the live API)
validate	provenance reminder
demo <name> / demos	example transcripts — /tritium/demos/

CLI — calls the same API: /tritium/cli/
Guide: /tritium/guide/