tern↗Get started↗

OPEN-SOURCE AGENT DISPATCH

Big ideas.
Lighter wings.

You bring the idea.
Your AI finds the right hands.

A lightweight skill for Codex. Work directly when it is simple, call Luna when it helps, and let the main AI check the result.

Start with Tern↗See how it decides↓
MIT LICENSEWORKS WITH PONYTAIL
THE FLIGHT BOARDInteractive demo
Tern seabird mark with open wings and a forked tail01 / UNDERSTAND03 / CHECK
ACTIVE AGENTS01
YOU
LUNALOW
✓ DONE

Clear rules. A lightweight agent handles the repeatable work.

BOUNDED CONTEXTHUMAN INTENT → CHECKED WORK
Small task? Stay solo.Clear scope. Fresh context.Dispatch. Check. Deliver.

01 / DELIBERATE BY DEFAULT

Just enough thinking.
Right where it matters.

Delegation starts with a decision.
Understand the goal. Then choose who does it.

Read the dispatch rules ↗
01

Small task? Stay solo.

A copy tweak or a quick answer needs no extra agent. Coordination has a cost, too.

MAIN
02

Clear scope. The right effort.

Luna Low for clear, repeatable work. Medium for connected tasks. Go deeper only when needed, or let the main AI take over.

LUNA
03

Dispatch. Check. Deliver.

Usually up to two agents at once; three only when useful. Bound their scope. The main AI checks, integrates, and delivers.

CHECK

02 / SHOW THE MATH

Does it save tokens?
Show the math.

More agents can mean more tokens.
A lighter model is not a lower token count.

Tern has no published benchmarks yet. Explore a hypothetical scenario, then compare the same task with and without it. These percentages are calculations, not promises.

Read the benchmark method ↗
TOKEN FLIGHT LOGIllustration · Not a benchmark
tokens

Same task, same acceptance criteria

tokens

Includes dispatch, context, review, and retries

tokens

Includes every agent, context, and retry

Fewer tokens in this scenario20.0%
Total with dispatch24,000Main AI + all subagents
Try a scenario

Formula: 1 − (main AI + all subagents) ÷ baseline. Count total input and output tokens; do not count cached input or reasoning tokens twice.

EFFICIENCY, WITHOUT THE FAIRY TALES.

Token usage, API cost, and subscription limits are different measures.Understand agent overhead ↗

03 / READY FOR TAKEOFF

Less coordination.
More creation.

Open source. Easy to read. Yours to adapt.
One small skill for the way you work.

TernMIT licensed · Made for Codex
01

Install in Codex

Run these two lines in your terminal.

# add the marketplace, then install Tern
codex plugin marketplace add https://github.com/iaredr/tern-dispatch
codex plugin add tern-dispatch@tern-dispatch
02

Bring your next idea

Start a new Codex task, select the skill, and ask.

›$tern-dispatchHelp me bring this idea to life.

Stays on for this conversation. Say “disable Tern” to stop. For project-wide activation, see the GitHub docs ↗.