Arus

Arus is waste-management software for multi-outlet hospitality in Indonesia. It gives a waste company, and the hotels and restaurants it serves, two systems on one dashboard: waste data recorded daily per bin and outlet, and separation audits run weekly on a tablet.

Its first operator is ACS Bali Foundation, an Indonesian waste company with hospitality clients across Bali. Digics designed and built the platform between July and October 2026, and it's now in pilot as ACS starts bringing its clients on.

Because one explains the other. Of the four streams Arus tracks, residue is the only one that can't be recycled, and how much residue a kitchen produces is decided by how well its staff separate. ACS's own argument, from the start of the design, was that a high residue share is the symptom and a poor separation score is the cause.

So the overview plots both on one chart: separation score against residue share, month by month. When the score climbs and residue falls, the connection is visible rather than asserted. When residue rises, the audits that explain it are one click away.

Each outlet's waste is weighed daily across four streams, each with its own destination:

• Organic food waste, which goes to a pig farm as feed.
• Organic compost, for composting.
• Non-organic, sorted for recovery.
• Residue, the part nothing can be done with, which goes to incineration.

Entering a day again replaces it rather than adding to it, so a day is only ever counted once. From those entries, Arus works out each outlet's composition and how much of its waste the waste company was able to recycle, at any period from a day to a year.

An inspector walks the outlet with a tablet, section by section and bin by bin. Each check gets one of two large answers, Achieved or Failed, sized for a thumb rather than a mouse. A failure opens a camera and a note, and the photo is required: no evidence, no failed check.

The score is the share of answered checks that passed, and the outlet passes when it clears its own pass mark with no critical failures. Pass marks are set per outlet, because a hotel kitchen and a beach bar don't fail in the same ways.

Each failure photo goes to Google Gemini for an independent read of what's in the bin. The model never sees the inspector's verdict, and it doesn't hand down one of its own: it reports what it detects, and the verdict is calculated in code from that. The review runs from a queue, off the inspector's screen, so an audit never waits on it.

Early testing showed why that design matters. In one evaluation photo, a model labelled styrofoam as rice. The fix was to make it describe the material it can see before it names the item, and to change the model; after that change it identified the styrofoam correctly in all four test runs. We'd rather an AI admit what it sees than guess what something is.

Every failure becomes a corrective action with an owner and a due date fixed seven days out, which is when the next weekly audit comes round. An action only closes with a photo of the fix, so "done" means the same thing to the waste company and the venue.

• Outstanding, overdue and closed actions are counted at the top of the board.
• Overdue actions surface on the overview next to the outlets that haven't logged their waste.
• Reports cover a day, a week, a month, a quarter, half a year or a year, and print to PDF for the client.

Arus is multi-tenant in four levels: the platform, the waste company, its client businesses, and their outlets. Each person's access says where they can act and with what authority, from read-only viewer to admin, so a hotel group's head office sees all its outlets while a venue supervisor sees one.

Each waste company runs Arus under its own brand, on its own address, with its own streams, colours, checklists and pass marks. ACS is the first, but another waste company, or a hospitality group running its own collection, is just another tenant. Partners can also connect through a REST API with keys, rate limits and sandbox accounts, still marked as unstable while it settles.

Arus is a Next.js application with Supabase underneath, hosted on Vercel, with the database and photo storage in Singapore, close to its users.

• App: Next.js and React, Tailwind CSS and Base UI, in a Turborepo with shared packages for the business rules and design tokens.
• Data: Supabase Postgres, queried with Kysely and plain SQL, plus Supabase Storage for audit photos.
• Auth: Better Auth, with roles and scopes checked on every query.
• AI: Google Gemini for photo review, run from a queue.
• Tests: more than 1,800 automated test cases, many of them running against a real Postgres database.

The pilot comes first: bringing ACS's hospitality clients onto the platform and learning from real audits rather than sample data. On the build side, the AI review queue needs to run on a schedule, the interface is English-only while many of the people using it day to day work in Indonesian, and a native tablet app with offline audits is planned but not built yet.

Arus is in pilot, so these are facts about how it's built and how it behaves, not usage figures.

Source: the Arus codebase and project history, November 2026. Screens on this page show the app's demo data, with fictional outlets.