Show HN: Delegate Claude Code's work to pi and a cheaper model

Claude Code plans and checks. A cheaper model does the heavy lifting.

Say "delegate to pi" and the heavy part of a coding task (reading files, editing, running tests) runs on pi, an open-source coding agent that can drive any model you pick, even one on your own machine. Claude only writes the brief and reads a short result, and your tests decide whether it worked.

Tip

On multi-file tasks, delegating cut Claude's cost by a third (−33%) with every hidden test still passing.

Note

The honest fine print. The numbers count Claude's cost only: pi's own spend comes on top (nothing if you self-host, cents on a hosted cheap model). Delegated runs are also slower, several times in our runs on a self-hosted pi (about 25 s plain vs 2-3 minutes). Samples are small (2 runs per task); the benchmark takes about five minutes to rerun yourself (how).

1. Add the plugin

/plugin marketplace add randomm/pi-delegate

/plugin install pi-delegate@pi-delegate

2. Set up pi once

npm install -g @earendil-works/pi-coding-agent # then run `pi`, /login (or export an API key), /modelCheap and local models work; the benchmark used a self-hosted Qwen. Also install jq, and on macOS

brew install coreutils for the timeout that bounds each pi call.

delegate to pi: add a

--jsonflag to cli.py, verify withpython3 -m unittest

Claude makes one call, pi does the work, and your verify command decides whether it worked:

EXIT CODE: 0

<pi's summary>

VERIFY: PASS (retries=0)

cli.py | 12 +++++++++---

If verification fails, pi gets one more attempt with the failure output. If it still fails, Claude says so instead of claiming success.

- A deterministic gate, not a second opinion. --verify "<your tests>"runs after pi. No model reviews the work: a same-model reviewer approves most changes, tests don't.

- Claude is told not to redo the work when the gate passes. Re-reading the diff and re-running the tests is exactly what ate the savings in our first benchmark.

- Guardrails: it refuses to run on your default branch or next to .env/*.pem/*.keyfiles, and disablesgit pushfor pi. This guards against mistakes, not a malicious model. For stronger isolation setPI_DELEGATE_WRAP(sandbox recipes in docs/configuration.md) or use a disposable clone or container.

skills/delegate/run.sh in this repo is plain bash. Any agent that can run a shell command can use it:

git clone https://github.com/randomm/pi-delegate.git

bash pi-delegate/skills/delegate/run.sh --verify "pytest -q" <<'TASK'

Fix the failing date parsing in utils.py; do not commit.

TASKbench/quick.sh -n 3 # about five minutes: plain Claude vs Claude + pi-delegate, prints a REWARD score

Method, tasks and the slower real-repo protocol: docs/benchmark.md · results. Settings (timeouts, safety, sandbox): docs/configuration.md.