Repair Relay

Try the interactive sample

https://hotragn.github.io/repair-relay-calle-hackathon/

No installation or API key required. These are fictional cases. Set capacity to one, read Maya's conversation, and accept her availability. Try adding Jo: the full session refuses a second proposal. Inspect Theo's correction and mark follow-up. No calls are placed and no appointments are booked in this sample.

Inspiration

Community repair workshops have limited places. A caller saying “Friday works—actually, I work Friday” creates a coordination problem: collecting a date is easier than deciding whether that date is supported by the conversation. Repair Relay focuses on that decision.

What it does

Repair Relay collects workshop availability through CALL-E and gives a coordinator the recipient's words to review before allocating a place. The workflow is: prepare candidate windows, approve the exact call request, retrieve the response, review the evidence, and export a handoff within session capacity. Availability is not an appointment confirmation.

How we built it

The application uses Python's standard library, SQLite, and plain JavaScript. CALL-E executes the phone task through its API. A connected local review desk reads saved results, records decisions, and exports the handoff.

The approval digest binds the recipient and request. Locally durable duplicate prevention protects against repeated dispatch, and an ambiguous creation response is not automatically retried. Refreshed results invalidate earlier reviews when their content changes. Exports require accepted responses, matching session times, and sufficient capacity.

What is verified

One authorized live CALL-E demonstration call completed on September 13, 2026 UTC. Request creation, result retrieval, evidence review, and a capacity-one handoff succeeded. The call disclosed that it was an AI demonstration for a fictional workshop. Private contact data and transcripts are not published.

Thirteen automated tests pass, covering transport handling, duplicate dispatch, stale approvals, malformed results, evidence checks, capacity enforcement, and the connected review desk. The organizer contribution passed repository validation.

Try it

Clone https://github.com/Hotragn/repair-relay-calle-hackathon and use Python 3.11 or later: python relay.py preview examples/case.json python relay.py evaluate examples/case.json examples/result.json python -m unittest -v python -m http.server 8766 --bind 127.0.0.1 --directory console/dist

Open http://127.0.0.1:8766. Review Maya's supported response, mark Theo's correction for follow-up, and set capacity to one before attempting to add a second proposal. Export the handoff. These sample conversations are synthetic.

For saved CALL-E results, the README documents the connected local review desk and opt-in live workflow.

What makes it useful

The reusable contribution handles the step between a phone response and a coordinator's decision: evidence review, changed-result handling, and capacity enforcement. Community repair is the specific initial use case. Its value to real coordinators still needs user validation.

Challenges and lessons

A completed phone task does not necessarily mean successful availability collection. A quote appearing in a transcript does not establish its meaning. Repair Relay therefore separates provider completion, evidence checks, coordinator acceptance, and handoff export.

Limitations and next steps

This prototype supports US recipients and English prompts, with one recipient per case. The video is a narrated overview; bundled examples are synthetic. Live verification covers one authorized call. We have not measured time savings or validated adoption with repair coordinators. Next steps are a coordinator pilot and a recorded walkthrough of the connected review workflow.

AI assistance

OpenAI Codex assisted with scoping, implementation, testing, documentation, and submission preparation. CALL-E performs the runtime phone task. AI assistance and synthetic examples are disclosed.

Community contribution

https://github.com/CALLE-AI/awesome-phone-call-agents/pull/507