Pre-pilot · Looking for first users
Akarsu brings INP and GIS import, automatic validation, scenarios, run comparison and reporting into one workspace. In English and Turkish.
Today
A water network model underpins design decisions, pressure management and investment plans. Yet it often sits on one computer, in a file called model_final_v3_really.inp.
Cleaning layers, mapping attributes and picking the right coordinate system starts from scratch on every project.
Which revision of the model, which input and which engine version produced a result is rarely on record.
Scenarios live in separate files, differences are copied into spreadsheets by hand, and a report starts ageing the day it is delivered.
Desktop software brings installs, licences and a dependency on one machine; a model is hard to share inside a team.
Every decision in the field rests on how right a model is.
Product
Akarsu keeps every step of modelling work in the browser, on the same model and the same record.
Start from your existing EPANET models or build a model from GIS data.
Data problems surface before a model is ever run.
Nothing is lost; every save is a revision.
Single-period and extended-period (EPS) simulations are solved by EPANET 2.2.
Scenarios are changes defined on top of a base revision.
Every deliverable rests on a traceable run.
Screenshots are from the real product, taken with the sample network on this page; the validation shot uses a copy with deliberately added errors.
Live demo
The sample network below is being solved right now, in your browser, by EPANET 2.2. Change the hour, close a pipe, raise a junction's demand and see the difference: a small rehearsal of Akarsu's scenario and comparison workflow.
In this scenario some junctions are cut off from every source; EPANET's results for them are not meaningful.
Click a pipe on the map, or pick it from the list in the Selected element tab, to close or reopen it.
First select a junction on the map or from the list in the Selected element tab.
Tap a junction or a pipe on the map, or choose one from the list above.
The sample network is synthetic; it does not represent any real town or utility. 133 junctions · 192 pipes · 1 pump · 1 tank · 1 pressure-reducing valve.
Download this network's INP fileEngineering
A hydraulic model is only worth as much as your trust in its results. Akarsu protects that trust by design.
Akarsu never computes or corrects a hydraulic result itself. Every pressure and flow comes from EPANET 2.2 and is stored unrounded, in the model's own units; screens round only for display.
A model stays in its own unit system; nothing is converted silently. All ten EPANET flow units are supported.
The engine's raw output cannot be changed. Derived values, such as the headloss gradient, are kept separate and labelled as derived.
Every run records the engine version and the SHA-256 digests of the uploaded file and of the exact engine input.
A model whose import recorded a parsing problem is not run as imported. Input that measurements showed EPANET crashes on, or reads differently from Akarsu, is refused before it reaches the engine; the known remaining cases are recorded as open defects.
Only rules and the engine produce validation findings and hydraulic results. AI stays outside those layers.
For the record: our reference benchmark set is limited to EPANET's Net1 example for now and is awaiting review by an independent hydraulic engineer.
AI
Akarsu's assistant answers questions in English or Turkish by reading stored model, validation and run data. Its language model is Anthropic's Claude. It is in preview: its evaluation with the live model has not been run yet.
Every value in an answer cites a numbered source and must match that stored value, in its unit; if one does not, the answer is withheld.
It does not edit, run simulations or write reports, and it never goes beyond the asking user's permissions.
It stays off in every project until the project owner turns it on.
AI clients that speak the Model Context Protocol get 11 read tools under the user's role: projects, models, scenarios, elements, validation, runs, results and comparison. They connect over stdio or Streamable HTTP with a credential bound to one user and one project.
Read tools
list_projectslist_modelsget_model_summaryget_elementsearch_elementsget_validationget_runlist_runslist_scenariosquery_resultscompare_runsChange proposals (propose_edits) are stored as pending proposals only; nothing is ever applied by itself.
Use with customer data opens only after a recorded evaluation with the live model and data-processing terms agreed with you.
Who it is for
Build the model for a water supply project, compare scenarios and deliver with a report in English or Turkish. Every deliverable rests on a traceable run.
Keep your network model in-house with role-based access (Owner, Editor, Viewer). Every model edit is saved as a revision with its author, so the model stays current for pressure management and water-loss work.
Teach and research hydraulic modelling with an EPANET-based workspace in English and Turkish.
Water network models hold the most valuable knowledge about how a city moves its water, yet they often stay on one computer, in one person's head. I am building Akarsu so these models can be open, traceable and shareable. I want to shape it with its first pilot users, around the real needs of engineers in Turkey.
FAQ
OWA EPANET 2.2.0. Akarsu does not write its own hydraulic solver; every hydraulic result comes from EPANET and is stored as is.
Yes. INP files are read as UTF-8 or as Windows Turkish (CP1254, which also reads ISO-8859-9 text). Sections that are not parsed reach the engine unchanged. A model can be exported back to INP; tests check that the export solves like the original, and the known exceptions are on record.
GeoJSON today. Points become junctions, tanks or reservoirs; lines become pipes. You declare the coordinate system (TUREF TM27–TM45, UTM 35–38N, WGS84). Shapefile and GeoPackage are not supported yet.
Pilot installations run on your organisation's own server: a single Docker image and PostgreSQL. If a project owner turns the assistant on, questions and the model data needed to answer them are sent to Anthropic. A hosted cloud service is not offered yet.
Projects are shared with Owner, Editor and Viewer roles. Every model save is a new revision; a save based on an out-of-date revision is refused with a conflict warning. Scenario overrides are saved separately and are not versioned. Real-time co-editing does not exist yet.
Water quality, fire flow, energy and transient analyses; SCADA connection and calibration; Shapefile and GeoPackage import; real-time co-editing and cloud hosting. We want to decide what comes first together with our first users.
Pricing has not been announced yet.
Pre-pilot. The core workflow (import, validation, editing, simulation, comparison and reports) works; we are looking for our first pilot users.
Tell us your organisation, your role and the rough size of your network, and we will set up a demo for you.
