The 45-Year History of Desktop Databases: From dBASE to AI Agents

Era 1: The dot prompt (1979–1984)

Before the graphical interface, a blinking period was the entire user experience. Type a command. Press Enter. Wait for the floppy drive to stop spinning.

In 1979, Wayne Ratliff wrote a database manager in 8080 assembly language on his IMSAI 8080 microcomputer running CP/M. Ratliff was a contractor at the Jet Propulsion Laboratory. He named the program Vulcan, after Mr. Spock's home planet. Its original purpose was to calculate football betting pools.

In 1981, marketers George Tate and Hal Lashlee licensed Vulcan and released it as dBASE II through their company, Ashton-Tate. They skipped "I" to make the software sound mature. dBASE II ran on CP/M and the IBM PC under MS-DOS, and it became one of the dominant business applications of the early PC era, alongside Lotus 1-2-3 and WordStar.

The interaction model was the "dot prompt": a single period on a blank screen. Users typed procedural commands like USE CUSTOMER, LIST FOR STATE = 'CA', and APPEND. Tables lived in flat binary .dbf files, with indexes stored separately in .ndx files. There was no GUI, no form layout, and no mouse. Everything fit in 64 KB of RAM.

dBASE spawned an aftermarket of third-party tools. Nantucket released Clipper in 1985, a compiler that turned dBASE code into standalone DOS executables. Developers could ship applications without requiring the customer to own dBASE. Fox Software, Arago, and others built compatible engines that extended or replaced the original.

Two other products from 1982 planted seeds that would matter later. Nashoba Systems released Nutshell, a text-mode MS-DOS database designed for non-programmers. Nutshell handled text indexing and basic forms. It was the direct structural ancestor of FileMaker. Across the Atlantic, Blyth Software shipped Omnis for the Apple II and Apple III, an early multi-user microcomputer database that later became one of the first cross-platform development environments.

Era 2: The form revolution (1985–1991)

Apple shipped the Macintosh in January 1984. Within a year, database software changed shape. Rows on a green-screen terminal gave way to canvases where users drew interactive, printable forms.

In 1985, Nashoba Systems rewrote their DOS Nutshell engine for the Macintosh. When their publisher, Leading Edge, declined to distribute the Mac version, Nashoba released it independently as FileMaker. The design introduced a concept that still defines desktop databases: the separation of data from layout. Users toggled between Browse Mode (entering records) and Layout Mode (drawing text labels, field boxes, and borders on a visual canvas). A single table could have a dozen distinct layouts — a phone directory, a customer card, a packing slip, an invoice — all from the same data.

Also in 1985, Ansa Software released Paradox for MS-DOS. Its innovation was Query by Example. Users placed checkmarks and typed sample criteria into an empty table grid. The engine translated these spatial hints into relational joins without SQL or procedural code. Borland acquired Ansa in 1987 for $29 million.

Apple formed Claris as a software subsidiary in 1987 and acquired Nashoba Systems, bringing FileMaker into the Apple product family. Claris released FileMaker II in 1988 and FileMaker Pro in 1990.

In France, Laurent Ribardiere developed a relational database for the Macintosh under the codename Silver Surfer. It shipped in 1987 as 4th Dimension (4D), one of the first desktop databases with a graphical schema editor and built-in client-server networking.

Meanwhile, David Fulton at Fox Software created FoxBASE, a dBASE-compatible engine that ran up to ten times faster through bytecode compilation. In 1989, Fox Software released FoxPro 1.0, which featured a multi-window text-based GUI with mouse support on MS-DOS — a graphical interface without a graphical operating system.

Era 3: The golden age (1992–2000)

The 1990s were the decade of rapid application development. Desktop databases became the engine behind small businesses, corporate departments, medical clinics, and schools. Three events in 1992 shaped everything that followed.

Access arrives at $99

After an earlier OS/2-based project codenamed Omega failed, Microsoft rebooted its desktop database effort under project Cirrus. In November 1992, Microsoft launched Access 1.0 at COMDEX Fall in Las Vegas. The introductory price was $99 — at a time when Paradox and dBASE retailed for $495 to $795. Access shipped over one million copies in its first ten months.

Access combined the Microsoft Jet Database Engine (a true relational engine supporting SQL), visual form and report builders, and Access Basic. It gave non-programmers a way to build working business applications with push buttons and dialog boxes.

Access 2.0, released in 1994, replaced Access Basic with Visual Basic for Applications. This aligned Access development with the macro language in Word and Excel. A community formed around the product: user groups, annual conferences, and an aftermarket of third-party add-ins for calendars, data grids, and replication.

Microsoft buys Fox Software

Also in 1992, Microsoft acquired Fox Software for approximately $175 million in stock. The purchase brought FoxPro's Rushmore query-optimization technology into the Microsoft stable. Rushmore techniques eventually found their way into the Jet engine and later into SQL Server.

FileMaker goes cross-platform

Claris brought FileMaker Pro 2.0 to Windows 3.1 the same year, making it the first major desktop database to run natively on both Mac and Windows.

The Office bundle effect

In 1995, Microsoft bundled Access into Office Professional. Millions of desktop computers suddenly had a relational database installed by default. Access 95 and Access 97 became the corporate workhorse for custom business workflows around the world. Access 97, with its Jet 3.5 engine, is still remembered as the high-water mark of desktop Access development.

FileMaker kept pace. FileMaker Pro 3.0, released in 1995, added true relational capabilities: multiple tables could relate to each other through key fields, and child records appeared in scrolling "portals" on a parent form. This was a leap from flat-file lookups to genuine relational design.

In 1998, Apple dissolved Claris as a software subsidiary and re-formed it as FileMaker, Inc., a single-product company. This narrower focus sustained FileMaker through several more major releases over the next two decades.

The fall of dBASE

Ashton-Tate's buggy release of dBASE IV in 1988 had already damaged the brand. Borland acquired Ashton-Tate in 1991 for $439 million in stock, but could not unify dBASE, Paradox, and its new Delphi environment while Microsoft undercut the market on price. By the late 1990s, dBASE and Paradox had faded from mainstream use.

What made it work

Golden-age desktop databases shared a few architectural traits. Everything — tables, queries, forms, reports, macros, and scripts — lived in a single compound file (.mdb, .fp3, .fp5). Multiple users connected to the same file over a local network, with record locks managed by temporary lock files. Report designers used banded layouts: page header, group header, detail row, group footer, page footer. And embedded scripting (VBA in Access, ScriptMaker in FileMaker) let non-programmers automate entire workflows.

Era 4: The Mac divergence (2001–2013)

As the 2000s began, Microsoft and Apple's database paths separated permanently.

Access stays on Windows

Throughout the 1990s and 2000s, Microsoft's Mac Business Unit ported Word, Excel, PowerPoint, and Outlook to macOS. They never ported Access. The reasons were architectural: Access depends on the Jet/ACE database engine, which was written in C++ around Win32 memory management and Windows file I/O locks. Its forms rely on COM, ActiveX, and OLE automation — subsystems that do not exist on macOS. Its VBA runtime calls directly into Windows DLLs and the Windows Registry. Porting Access would have meant rewriting the entire engine from scratch. For a deeper look at why Microsoft never shipped Access for macOS, we covered the technical details separately.

For why there is no Access for Mac, the short answer is that the cost of rewriting exceeded the revenue it would generate — and keeping Access on Windows gave Microsoft a reason for enterprise workgroups to stay on Windows PCs.

FileMaker 7 and the Relationship Graph

In 2004, FileMaker Pro 7 arrived as a major rewrite. It unified multiple tables into a single .fp7 file and introduced the Relationship Graph: a visual canvas where developers dragged lines between table occurrences to define joins, cascade deletes, and relationship rules. The structure of the database was visible at a glance on a single canvas.

Bento: the $49 personal database

In January 2008, FileMaker Inc. launched Bento, a $49 consumer database built from scratch for Mac OS X Leopard. Bento rejected relational complexity in favor of a simple, visual interface. It offered drag-and-drop card trays, customizable themes, and zero-configuration sync with Apple Address Book and iCal. In 2010, Bento launched on iPad and became one of the most popular early iPad database apps.

Bento confirmed what the industry had forgotten: card-and-tray layouts made non-programmers willing to use structured records.

Bento dies

On September 30, 2013, FileMaker Inc. discontinued Bento. The most common explanation is that Bento cannibalized FileMaker Pro sales. At $49, small business operators, photographers, event planners, and consultants were buying Bento instead of FileMaker Pro, which cost $299 to $499. FileMaker Inc. never confirmed the reason publicly.

The decision stranded tens of thousands of users without a modern personal database on the Mac — a gap that remains open more than a decade later.

Era 5: The cloud shift (2014–2023)

With desktop databases abandoned by Apple and priced for enterprise by Microsoft and Claris, venture capital funded a migration to the browser.

Airtable and Notion

Airtable, founded in 2012 by Howie Liu, Andrew Ofstad, and Emmett Nicholas, reimagined the database as a colorful cloud spreadsheet with typed columns — attachments, checkboxes, single-select pills — and relational links between tables. Notion added inline database blocks (tables, boards, galleries) to its wiki workspace in 2018, bringing lightweight data storage to millions of students and knowledge workers.

Platforms like Ninox, SmartSuite, and Coda followed. All were web-based, all charged per seat, and all required an internet connection.

What went wrong

Cloud databases made collaboration easy. They also introduced trade-offs that accumulated over a decade.

The form disappeared. Airtable and Notion prioritized the grid view. Users scrolled through wide horizontal tables. The focused form layouts that FileMaker and Bento had refined over twenty years were gone. (We wrote about the death of the database form separately.)

Per-seat pricing compounded. Airtable grew from a consumer tool into an enterprise sales operation, raising prices to $20–$45 per user per month. Automation runs were capped. Prorated refunds disappeared. For a solo operator or small team, the annual cost of a cloud database subscription exceeded what a perpetual desktop license had ever cost.

Data moved to someone else's server. Records, customer lists, and business files lived behind proprietary cloud APIs. When the Wi-Fi dropped or AWS had an outage, work stopped. Exporting was possible but never complete — views, automations, and linked-record logic stayed behind.

Desktop holdouts went subscription too. Even Tap Forms, one of the last offline Mac databases modeled after Bento's templates, transitioned to recurring subscription pricing.

FileMaker followed. FileMaker, Inc. rebranded as Claris International in 2019 and moved FileMaker to subscription pricing at $19 per user per month. The product that had once served solo operators now targeted team deployments and enterprise integrations.

Era 6: The local-first era (2024–2026)

Two problems converged in the mid-2020s. One was about humans. The other was about machines.

The agent problem

Between 2023 and 2025, developers tried to give AI agents memory by storing facts in loose markdown files — Obsidian vaults, text directories, chat logs. The results were poor. Agents hallucinated field names (client_email versus email versus contact), failed at relational aggregation ("sum all outstanding invoices for Acme Corp"), and burned context-window tokens on irrelevant prose. For more on why agents need databases, we covered the technical argument in detail.

The fix was already decades old. Agents need the same thing that businesses needed in 1992: a typed schema, foreign key constraints, indexed lookups, and transactional writes. They need a relational database.

The operator problem

Developers responded by telling users to "just use SQLite" or "just use Postgres." Solo business owners, consultants, and operations leads will not write raw SQL in a terminal. They need what they have always needed: forms, cards, status badges, and printable layouts.

The synthesis

The answer is a desktop database that serves both audiences through the same file. SQLite stores the records. A TOML schema defines the tables, fields, and views in a human-readable text file. The desktop app renders forms and cards for the operator. An embedded Model Context Protocol server exposes the schema and records to AI agents, which can read, query, and write data — with every change routed through the operator's approval before it commits.

SQLite, written by D. Richard Hipp in 2000, is the most widely deployed database engine in the world. It runs on every smartphone, every web browser, and most embedded devices. A SQLite database is a single file. It needs no server process, no configuration, and no network connection. These properties make it the natural storage engine for a local-first desktop database.

A project is a folder on your disk. It works offline. It responds instantly. You can back it up by copying the folder. You can inspect the schema in any text editor. An agent can use it as a local client and customer database without touching a cloud API.

Over forty-five years, the interface moved from the dot prompt to the form to the grid to the cloud. The architecture is now returning to local files, typed schemas, visual forms, and a structured interface for agents.

Timeline

What comes next

Dosirak is a local-first desktop database: a folder on your disk with a TOML schema, SQLite records, forms and cards for the operator, and an embedded MCP server for agents. See what Dosirak does.

Questions