Muhamed Isabegovic
All projects

Coming soon

tabula

Understand a system you inherited, write down what it should become, and hold every agent to it.

Role
Author
What
One Rust binary that keeps a project's memory, contract and judge as plain files
For
Menders: people who look after code they did not design, and the agents beside them
Works with
Any agent harness: the command line, MCP, the canvas or your editor; small models included
Status
In private development
tabula, in private development: understand a system you inherited, write down what it should become, and hold every agent to it.

Most tools are built for the architect. Most of us inherit.

Most tools are built for the person who draws a system before it exists. Most of us look after one someone else built, full of decisions nobody wrote down, and agents will happily change it without knowing any of them.

tabula helps you understand the system you inherited, write down what it should become, and hold every agent that works on the code to what you wrote.

Everything tabula will do.

The full plan, theme by theme. Working today means built and in use, though not yet released.

  • Working today

    Your files are the truth, shared by many agents.

    Pages are Markdown in git. Sync commits, fetches and merges, and on a conflict it names the page and pushes nothing. Several agents share one brain without a server: each claims work on a lease, and none can release another’s claim. What an agent reports goes into a journal no command rewrites, and findings wait in an inbox for a person’s verdict. One level above every repository covers work that spans them.

  • Working today

    Intent before any code moves.

    An entry gate refuses an edit until the work is aligned with the plan or decision it comes from, and work no page governs asks for new intent a person accepts. Effects record their intent before they run. Next: every sentence carries its standing, proposed, accepted or contested, and any file can answer why it is the way it is.

  • Working today

    Judged by something other than the author.

    evaluate settles each acceptance line, and each premise of a proposal, without asking the agent that did the work. Every verdict carries a receipt that can be checked again, and a run’s whole story can be rebuilt from the journal. Mutation runs show which tests catch nothing. Next: every judge proves it rejects answers known to be wrong.

  • Working today

    Draw your architecture on a canvas.

    tabula ui is a canvas served by the tabula binary on your machine: typed parts (module, service, database, table, endpoint, queue, job, external) and arrows, zoom through the C4 levels from context to code, a part expanded in place. The same picture prints in the terminal, renders as SVG, and exports to Mermaid, draw.io or picjs. A change is drawn on the picture: what it did since the baseline.

  • Working today

    Drawings that compile into enforced rules.

    A “never” arrow, or a law written as a sentence, compiles into enola constraints the repository’s own CI checks, with your reason beside each rule. distance names exactly where the code breaks a rule today, by file and line, and writes a step-by-step brief to build what is planned. Exceptions are signed, never silent. Planned: a loop that works the gap until it closes.

  • Working today

    Explain a codebase you inherited.

    A tour walks a newcomer through it one step at a time. explain writes a page a newcomer can read, from facts the engine measured; ask answers with file and line; trace follows a path between two functions with the tests and decisions on each step. Where a decision came from, the likely causes of a symptom, and the files both complex and always changing, each one command away.

  • In progress

    A library of every known architecture.

    Styles, patterns and practices, with their sources and licences, kept as pages. tabula says whether your code keeps the style it claims and what that style trades, in measured numbers; adopting one turns it into the laws agents are held to, and answers cite the library rather than a model’s memory. The aim is the whole history of the field.

  • Working today

    Every door, every verb naming the next.

    The same verbs from the command line, over MCP, on the canvas and in your editor through a language server. tabula installs into Claude Code, pi, Codex, Amp, Goose and OpenCode, and every gesture on the canvas is also a verb an agent can call. Every answer ends with the command to run next and why.

  • In progress

    Built for small and local models.

    Handles, outlines and single sections instead of whole pages, a compact profile, and a cap on every result. Self-hosted models are first-class, and tabula scores how well a model uses its tools. A small model is the measure of its efficiency, never a requirement. Planned: context chosen by relevance, ranked on your machine with nothing sent anywhere.

  • Working today

    Workflows, schedules, CI and docs that run.

    Workflows are pages: each step goes to its owner, a run keeps the checkpoints it started under, and runs are drawn from the journal. Scheduled jobs, tabula ci and sketches through whatever harness you use. The documentation is a tabula brain whose commands run in the test suite. Planned: the documentation published as its own site.

Coming soon

Screens and real runs, soon.

tabula is in private development, and much of the plan above already works on my own projects. Before it opens, its documentation goes online, the binary ships with checksums, and a first reader works through the homepage and a tutorial. This page will then show the canvas and real runs, as the other tools’ pages do.

If you look after a system you inherited and want to try it early, write to me.

What it shows about how I work.

tabula is what my knowledge-base practice becomes when it is one tool: the decisions written down, the code held to them, and the people kept in charge of what is true. I build it the way I want teams to work with agents, and I run my own work on it.

Inherited a system?

Tell me what you look after and how your team works with agents, and I will tell you honestly what I would set up first.

Get in touch

Want a second opinion?

Start with an architecture review. Fixed scope, one to two weeks, and a plan your team can act on.

About the review

Or email info@muhamed.at