@trama-dev/cli

Agents don't need frameworks. They need a runtime.

The CLI for trama — the agentic runtime that turns intent into self-contained, shareable agent programs.

npm install -g @trama-dev/cli

Five commands. That's the whole interface.

trama create optimizer "improve sort.ts by benchmarking alternatives"
trama run optimizer
trama update optimizer "also track memory usage"
trama list
trama logs optimizer

create generates a complete agent program from one sentence. run executes it — with auto-repair if it crashes. update rewrites it with new requirements. The output is real TypeScript you can read, diff, and share. Anyone can git clone && trama run.

What trama generates

From a single sentence, trama produces programs that range from one-shot pipelines to autonomous optimization loops to self-orchestrating multi-agent workflows — where trama programs create and run other trama programs.

# One-shot pipeline
trama create hn-digest "fetch the HN API, get top 10 stories, write digest.md"

# Autonomous optimization (autoresearch pattern)
trama create sort-opt "benchmark sort.ts, propose improvements, keep only faster versions"

# Self-orchestration — trama composes trama
trama create research "break topic into sub-questions, create a trama sub-program for each, \
  run them, synthesize a final report"

CLI options

trama create <name> <prompt> [--model <model>] [--arg key=value ...]
trama run <name> [--timeout <ms>] [--arg key=value ...]
trama update <name> <prompt>
trama list
trama logs <name>

See the main README for the full picture — why a runtime matters, what generated programs look like, and the runtime API.