AvenfileA customer-records workspace for independent service businesses, on the desktop and in the browser.

Avenfile is where the owner of a small service business finds a customer, reads what happened last and records the next step. It runs as a native desktop app with an encrypted database on the computer, and in the browser as a private sample workspace.

Avenfile is an Awan Forge product, designed and built by Awan Forge. The desktop app is downloaded from its product page, and the browser app opens a sample workspace for trying it.

Open the sample workspace

A private workspace with sample customers. No account.

Browser

Avenfile in the browser: the Customers list with saved views, search, Filter and Sort above a table of customers, and the current customer, Bethan Greenwood, shown in an inspector on the right with contact details, next follow-up, tags and timeline.
The Customers screen, captured from the live sample workspace on 11 November 2026. The current customer is marked by a bar in the list and shown in the inspector on the right: contact details, next follow-up, tags and timeline.
Type
Awan Forge product
Platforms
Linux desktop, web app
Our work
Research, interface design, shared C# core, desktop app, web app and API, testing, packaging, deployment
Built with
.NET 10, Avalonia, SQLite, ASP.NET Core, React
Date
November 2026

Find a customer, read what happened last, record the next step

An independent service business keeps its own customer records: who the customer is, how to reach them, what was done and what is due next. Avenfile is built around that loop for one person, and the design serves it and nothing else.

We wrote the essential requirements first, then the four things that make the product worth choosing.

Essential

  1. Easy navigation
  2. Secure customer storage
  3. Search and filtering
  4. Add, edit and delete customers
  5. Backup and restore

What Avenfile adds

Fast retrieval
One search field across name, business, every email address, every phone number and tags.
Customer timelines
The application writes an entry when something happens. Nothing is generated to fill it.
Follow-ups
Due today, overdue and upcoming, with counts computed from the records each time they show.
Recovery that can be trusted
A restore shows what will change, saves a recovery copy first and replaces the data in one transaction.

The first layout, drawn before any interface code

On 11 November 2026 at about 07:53, before the design rules, the tokens or any application code existed, we drew the five main screens as grey boxes in plain HTML and captured them at 1440 and 390 px wide. The files are kept read-only as the record of where the design started. The names in them are placeholders.

The first layout already settled the structure: a sidebar, a list with the current customer in a panel beside it, a separate full profile, and import and restore as steps with a review before anything is written.

First wireframe

Grey-box wireframe of the Customers screen: a sidebar, a row of four saved-view buttons, a search field with Status, Tag, Location, Last contact and Sort buttons on a second row, a table of placeholder customers and an inspector with boxed sections.
Customers at 1440 px, as first drawn. Saved views and the search row are two separate toolbars, and the four filters are always on screen.

Desktop app, Ubuntu

The final Customers screen in the desktop app: saved views, search, Filter and Sort in one row above the customer table, the current row marked with a bar, and an inspector without boxes on the right.
The same screen in the desktop app. One toolbar row, filters behind Filter, the inspector divided by lines, and the storage state in the sidebar foot.

What the first wireframe got wrong

  • Two toolbar rows became one

    Saved views sat on one row, and search with four filter dropdowns on another, all visible at all times. The final design puts them in one 48 px row with the filters behind Filter, which gives the list one more row.

  • The inspector lost its boxes

    Every inspector section had its own outline, so the panel looked heavier than the list it serves. Sections are now separated by space and a single line.

  • Where the data lives moved to the sidebar

    Sample workspace sat in the header beside the actions. The sidebar foot now always names where the data is: Saved on this computer, or Sample workspace.

  • The phone layout stopped clipping

    At 390 px the Filters button was cut off at the right edge, and the bottom bar's middle item was covered. The narrow layout gives Filter and Sort a row of their own.

  • Failures appear where they happen

    The backup page listed failure cases in a static box. In the final design each message appears beside the step that failed.

  • Follow-ups open on what is late

    The tabs showed counts without saying which opens first. The screen opens on Overdue whenever it has any.

First wireframe

Grey-box wireframe of the Customers list at 390 px wide, with the Filters button cut off at the right edge.
390 px, first drawn. Filters is clipped at the edge.

Browser

The Customers list in the browser at 390 px wide: search, saved views that scroll sideways, Filter and Sort on their own row, two-line rows and a bottom bar with Customers, Follow-ups and More.
390 px in the browser. Filter and Sort have their own row.

Six design rules, each with a source and a check

Before implementation we wrote six principles, each from a source we read, turned into rules for this product and into a check a tester can run. One check per principle is shown here.

Design principles, their sources and one acceptance check each
PrincipleSourceOne of its checks
One quiet frame, content firstSaarinen and Gillet, “How we redesigned the Linear UI (part II)”, Linear, 2024On Customers at 1440 x 900 with one row current, exactly three elements use the accent as a fill: Add customer, the navigation marker and the row marker.
Show the common things, disclose the rest on requestJakob Nielsen, “Progressive Disclosure”, Nielsen Norman Group, 2006With no filter set, the toolbar holds exactly three controls besides the saved views.
The system always says what state it is inNielsen's heuristic 1, Visibility of System StatusImport a file of N rows: imported, skipped and with problems add up to N.
Recognise, do not recallNielsen's heuristic 6, Recognition Rather than RecallSearch, set two filters, sort, open the 30th row's profile and go back: all of it is unchanged and the row is in view.
Prevent the mistake, and make the rest reversibleNielsen's heuristic 5, Error PreventionA restore with a wrong passphrase leaves the customer count identical, and the message ends with “Nothing was changed.”
Everything works from the keyboard and is readableW3C, WCAG 2.2Muted text measures at least 4.5:1 on canvas, surface, row hover and the current row (computed 5.00, 5.45, 4.93 and 4.66).

A change made after looking at the real screen

At 1440 px with the inspector open, the list is about 816 px wide. With six columns, customer names and tags were cut short. Below 900 px of list width, the business now moves under the customer name, in both apps, and the Business column is not drawn.

816 px

list width at 1440 px with the inspector open

900 px

list width below which the business moves under the name

Browser

Six rows of the customer list. Where a customer has a business, it appears in a smaller line under the customer's name, and the list has no separate Business column.
Rows at 1440 px with the inspector open. The business is the second line under the name, in the browser and in the desktop app.

What we borrowed, and what we left

We opened three interface references and two motion references in a real browser window, and wrote down for each what Avenfile takes as an idea and what it leaves alone.

References, what we borrowed and what we left out
ReferenceBorrowedLeft out
Behance, “CRM Client Management & Billing Dashboard” by Monil SoniThe list comes first; search and filters share one toolbar row; status is a dot plus a word; the list says how many customers it shows.Pagination, which would lose the current row and scroll position; an eight-step create flow for a small record; billing columns.
Attio, public product, help and changelog pagesA record opens beside the list and a full page is one step further, with one structure for both; saved views at the top left; history as a timeline of things that happened.Configurable objects and columns, enrichment, generated summaries, email and calendar sync.
Linear, “How we redesigned the Linear UI (part II)”The sidebar and header as one quiet frame; every sidebar icon, label and count on one grid; the accent kept out of large surfaces.Header tabs, a custom theme generator, and the density of a tool for all-day keyboard users.
HoneyQuote, motion studyThe browser's own scrolling, content readable at once, control feedback no longer than 200 to 270 ms.Transitions on every property, and a slow-start curve for feedback.
Clinia, motion studyOne product walkthrough with numbered, labelled steps, and a reduced-motion path that removes entrance animation.A script that remaps the wheel and pins the page, a first second and a half with nothing to read, endless decoration.

Motion with three jobs

Motion confirms that input was received, shows where something came from, and marks a change of step. Feedback lasts 120 to 220 ms and starts fast. The only movement is 8 px, when the inspector first appears; moving between rows changes its content with a short fade. With reduced motion requested, every view swaps at once, and the interface works the same with all motion switched off.

Not inspected

  • The signed-in Attio and Linear applications. Everything about them comes from public pages.
  • Awwwards galleries, used only to find sites. A listing there is not treated as proof of anything.
  • w3.org in the scripted browser, which returned status 403. The WCAG text was read through a fetch and the criterion source files.
  • Behance projects module by module. Each was viewed as a first screen and four or five scrolled frames, two at reduced size.
  • Hover animations on the motion references. Declared styles were read; no hover change was seen.

Finding a customer, and keeping your place

From the keyboard, a customer that three letters identify is found and opened in six key presses: the slash key, three letters, Down and Enter. Going back to the list shows it exactly as it was left.

Search
One field matches name, business, every email address, every phone number and tags. Phone matching ignores spaces and punctuation, so 0207946 finds 020 7946 0018. An older answer never overwrites a newer one.
Saved views
Four fixed tabs, always visible, each with its count: All customers, Needs follow-up, Recently added and Archived.
Filters
Status, tag, location and last contact, behind Filter. Each filter that is set becomes a chip that names its value and can be removed.
Inspector and profile
A click or an arrow key makes a row current and shows it beside the list. Enter opens the full profile with every stored field.
Timeline
Notes, edits, imports, archiving and completed follow-ups, written when they happen, with their date and time.
Follow-ups
Today, Overdue and Upcoming. Completing one writes a timeline entry, sets last contact to today and offers Undo for 8 seconds.
Where the list remembers its state
StateBrowserDesktop
Saved view, search, filters, sort, current customerIn the URL, so refresh, Back, Forward and shared links restore themKept for the session; saved view and sort also survive a restart
Scroll position of the listSession storage, restored after a refresh and on return from a profileRestored on return from a profile

Desktop app, Ubuntu

Customer profile for Bethan Greenwood in the desktop app: contact methods with Copy and Open buttons, address, tags and background on the left; the next follow-up, a note field and the timeline on the right.
The profile: everything stored about one customer, with the next follow-up and the timeline beside the details.

Desktop app, Ubuntu

Follow-ups screen in the desktop app on the Overdue tab: follow-ups grouped under date headings, each with a checkbox, the action, the customer, the due date marked overdue, and Reschedule.
Follow-ups, opened on Overdue because it has entries. Each date says how many days late it is, in words as well as colour.

One C# core, and one SQLite file per workspace

Domain rules, use cases and storage live in a shared C# core. The Avalonia desktop app and the ASP.NET Core API both reference it and never each other, so validation, search, import, backup and restore are the same code in both apps.

A workspace is a single SQLite file with one schema. No table carries a tenant column, so no query can reach another workspace by forgetting a filter, and deleting a workspace deletes one file. We rejected one shared database with a workspace column on every table: every query would need a filter, and expiry would be a delete across many tables.

Shared C# core

  1. Domain
  2. Application
  3. Infrastructure

Validation, search, duplicates, CSV import, backup, restore, follow-up counts

Avenfile.Desktop, Avalonia

workspace.db on this computer, encrypted

Avenfile.Api, ASP.NET Core

One .db file per sample workspace

React app in the browser calls the API with a same-origin session cookie

Encrypted .avenfile-backup file, written and read by the same code, moves both ways

Both apps run the same core. Data moves between a desktop workspace and a browser workspace only in a backup file that the person exports and imports.

On the desktop

The workspace file is encrypted with SQLite3 Multiple Ciphers under a random 256-bit key. On Linux the key is held by the system keyring, or in a key file only its owner can read when no keyring answers. The Windows build protects it with Windows DPAPI.

In the browser

Opening the app gives each visitor a private workspace file, found from an HttpOnly session cookie. The server keeps only a hash of the token. No endpoint accepts a workspace id: the workspace is resolved from the session for every request, including export, backup and restore. There is no sync between the two.

Import, backup and restore: review first, then write

Import never merges customers. Its review step lists the rows that are ready, possible duplicates by email address or phone number, and rows with problems, before anything is written.

Desktop app, Ubuntu

Import review step in the desktop app: counts for ready rows, possible duplicates and rows with problems, a table of problem rows with the reason for each, the choice to skip or import duplicates, and the button Import 4 customers.
The review step of a CSV import, with each problem row and its reason. In the desktop run, a file of 9 rows gave 4 imported, 1 skipped and 4 with problems.

The backup file

A backup is one file: the word AVENFILE, a format version, a header and the ciphertext. The key comes from the person's passphrase, at least 12 characters, through PBKDF2-HMAC-SHA256 at 600,000 iterations with a random salt, and the data is sealed with AES-256-GCM. The passphrase is never stored.

Inside is a logical export of the customers, contact methods, tags, timeline and follow-ups. It carries no session token, workspace id or server path, so the same file restores on the desktop and in the browser. A new backup is decrypted and checked again before it is written.

600,000

PBKDF2-HMAC-SHA256 iterations

256-bit

AES-GCM key

Desktop app, Ubuntu

Backup and restore in the desktop app: on the left a backup created with its file name, counts, size and a passed integrity check; on the right the restore review comparing counts in this workspace and in the backup, with Cancel and Replace current data.
A backup created and its integrity check passed; on the right, the review of what a restore will replace. The confirm button is named after its effect.

Restore, staged

  1. Read the header, and refuse an unknown format or version.
  2. Derive the key and decrypt. A wrong passphrase or a damaged file fails the AES-GCM tag check.
  3. Decompress with a size cap, check the schema version and the hash of the data, and check every record with the same rules as typing it in.
  4. Show what will change: the backup's date and counts beside the current workspace.
  5. On confirmation, save a recovery copy with SQLite's online backup, then replace the data in one transaction.

When it fails

Wrong passphrase, or a changed byte
“The passphrase is incorrect or the file is damaged.” Nothing changes.
A truncated file, or one that is not an Avenfile backup
Refused. Nothing changes.
A format version this build does not know
Refused with a clear message. Nothing changes.
The replacing transaction fails
It rolls back, and the recovery copy remains.

Desktop app, Ubuntu

Restore step in the desktop app after a wrong passphrase: File format Passed, Passphrase and encryption Failed, and the message This backup could not be opened. The passphrase is not correct, or the file is damaged. Check the passphrase and try again. Nothing was changed.
A wrong passphrase. The checks say which part failed, and the message ends with “Nothing was changed.”

What we ran, on 11 November 2026

51/51

automated .NET tests passed

3.21 ms

per search, averaged over 50 warm runs (conditions below)

Checks run on 11 November 2026 and their results
CheckWhat it coveredResult
Automated .NET tests on Ubuntu 24.0428 core, 19 API and 4 desktop headless tests on real SQLite files: CSV with malformed and duplicate rows, a backup round trip, a wrong passphrase, a flipped byte, a truncated file and a bad header each leaving the data unchanged, and two sessions kept apart across export, backup, restore and recovery copies.51 of 51 passed
Browser journey against the production build and a real APIAdd, search, edit, archive and undo, export, back up, a wrong passphrase refused, restore, reload; and two visitors given separate workspaces.2 of 2 passed
Live site, checked from outsidePages and security headers, and the same API journey in a throwaway sample workspace that was then deleted.Passed
Desktop app in its real window, Ubuntu 24.04 on a virtual displayFirst run, search, add with validation, archive and delete with undo, completing a follow-up, CSV import with malformed and duplicate rows, a backup through the save dialog, a wrong passphrase refused, restore, and a restart with the data kept.Passed
Linux packagesThe packaged build was started from its archive. The .deb package was installed on Ubuntu 24.04 and the installed app started.Passed
Accessibility, automatedaxe-core with the WCAG 2.2 AA rules on 10 web screens in both themes. Contrast ratios for every text and background pair are computed in the design document.No violations

Search was measured once: 3.21 ms per query averaged over 50 warm runs on the 100-customer sample set, for a search plus a location filter plus a tag filter, in a Debug build.

Known limits

  • The Windows package is cross-built on Linux and published; it has not been run on Windows yet, so the Windows key store, file dialogs and window are unverified there.
  • Not tested on real phones, in Safari or Firefox, or with screen readers.
  • The browser workspace is for sample data. It is deleted 7 days after its last use, and it is not encrypted at rest on the server: access is limited to the service account.
  • The desktop packages are unsigned.
  • The API runs on a single host, and SQLite allows one writer at a time per workspace file.
  • With a locked system keyring, the desktop error names a missing key file, which is misleading.

How we work

Avenfile shows the order we work in: requirements and a first layout before any code, design rules that each come with a check, one core shared by both apps, and tests that run against real files. What we have not tested yet is written down beside what we have.

Open the sample workspace

Sample customers in a private workspace, deleted 7 days after your last visit.