Allfedo

Allfedo helps known kitchens and shelters agree food pickups, handle changed plans and confirm what actually arrived. A local coordinator connects current food with today's receiving needs and an approved collector.

From surplus to a real handoff

A kitchen posts a counted food lot, packaging, allergens, holding conditions and collection window. Receiving teams confirm their total current need and receiving requirements. The coordinator asks the AI Hub agent for a suggested split, reviews it and sends invitations separately.

Sending creates temporary holds. Each receiving team accepts an exact quantity. An assigned collector then agrees the actual load, places and times. The kitchen records what physically leaves; receiving staff count what they accept. An email or planned pickup never becomes a delivered meal by assumption.

Actual app screen with fictional test data and a checked manual plan.

Keep the case useful when something changes

Partial pickups and receipts leave specific follow-up. A smaller promise needs current partner confirmations. Late reports and corrections preserve earlier evidence. Missing food does not automatically return to kitchen stock or create replacement demand.

If a receiving point is unavailable after collection, the collector reports the food still carried. A new receiving team accepts and the collector agrees the onward trip. Linked records keep the original pickup and account for the whole load. Later returns or disposal of rejected food follow the same boxes without counting them twice. Returned food remains quarantined.

Native InterSystems workflow

An InterSystems IRIS Production, the native integration workflow, routes signed-in commands through a Business Service and Business Process. Its native AI Hub agent uses OpenAI to propose a split or ask a concrete question. Its tools read only authorized food and receiving facts. The agent cannot reserve food or approve a pickup.

Embedded Python checks proposals and reconciles physical stock inside IRIS. Current shared quantities and evidence are checked again in the transaction that saves invitations. Failed emails stay in a saved queue and retry a limited number of times through the native email operation. Local demonstration messages go to Mailpit, a test inbox.

Try it

The tested runtime is AI Hub Community 2027.1.0AI.109.0, using gpt-4.1-mini. Download the licensed Community image from the official AI Hub portal; the repository does not redistribute it. Docker Compose and Python prepare the app, with six private fictional accounts for a walkthrough. Add your own OpenAI key for the live agent. Checked manual planning remains available without a key.

• Installation and source
• Try a complete collection
• Implementation and responsibilities
• Actual validation and measured agent usage

A fresh installation with new data volumes passed the native checks and complete five-account browser journey. Two paid native planning decisions ran separately during development. Ordinary automated tests use manual plans and make no paid model calls.

All demo partners and physical events are fictional. This release serves whole meal boxes and known partners; it has no public driver marketplace, route optimization or automatic food-safety certification. It has not been field-validated. Source uses the MIT license.