Tukun.ai 0.12 Is Live: One Workbench, Two Ways to Work

Tukun.ai 0.12 is now live.

This release does one main thing: it turns General assistant and Data analysis into two real ways of working inside the same workbench.

Not every task should start with a file upload, a database connection, or a governed analysis flow. Many tasks start earlier than that. The user needs to sort out the question, shape the review, or draft the first version of something.

Other tasks should start from data on day one. 0.12 makes that split explicit. You can begin in General assistant, then move into Data analysis when the work needs files, databases, governed semantics, and evidence-backed answers.

What 0.12 changes

The important change in 0.12 is not surface polish. It is that the workbench now treats two kinds of work differently on purpose:

• General assistant is for discussion, planning, drafting, and problem framing
• Data analysis is for governed analysis grounded in files, databases, and reusable evidence
• each mode keeps its own context instead of leaking setup and assumptions across unrelated turns
• setup actions like upload and data-source configuration now belong only to the data-analysis path

That makes the product easier to read. Not every prompt has to pretend to be a data question, and not every conversation has to carry data-analysis setup before it needs to.

Two modes inside the same workbench

Before 0.12, the workbench could technically support different behaviors, but too many tasks still looked like one generic flow.

That created a mismatch. Some tasks begin with questions like:

• what exactly are we trying to decide?
• how should this review be structured?
• which metrics matter before we query anything?

Other tasks begin with a file, a connected database, or a question that should stay tied to visible sources from the start.

In 0.12, the product acknowledges that these are different starting points. The same workbench can now begin as General assistant or Data analysis, without forcing both into one ambiguous mode.

Context now stays separate by mode

General assistant and Data analysis now keep separate context state. That means switching between them does not silently mix together different assumptions, data attachments, and task history.

In practice:

• a planning conversation in General assistant does not automatically drag data-runtime context into later turns
• uploads, connected databases, and source configuration stay attached to Data analysis
• when you return to a mode, its own context can continue from where that mode left off

This matters because context pollution is not a cosmetic problem. It changes what the system is allowed to assume and how trustworthy the next answer feels.

General assistant is not tied to one model vendor

There is another practical reason General assistant matters: it is not a thin wrapper around one model vendor’s own assistant product.

Many AI assistants still live inside one vendor’s stack. When you use them, you are usually also accepting that vendor’s model boundary, product assumptions, and upgrade path.

Tukun’s General assistant is different. It is a work mode inside the workbench, not another name for one specific model. You can switch between different models based on the task instead of being locked into one vendor’s assistant experience from the start.

That matters most for drafting, discussion, planning, and synthesis. These tasks often do not need data in the first turn, but they do depend on model style, reasoning behavior, and response quality. Separating the assistant surface from a single model vendor gives the user more room to choose.

Data setup is now clearly attached to Data analysis

0.12 also makes setup more explicit.

Uploading files, configuring data sources, and starting source-aware analysis are now treated as capabilities of Data analysis, not generic controls that appear equally relevant everywhere.

That sounds small, but it does clean up the product:

• General assistant no longer carries setup actions that it does not actually use
• Data analysis becomes more legible as the place where governed inputs, evidence boundaries, and analytical follow-up really happen

Not every surface should try to do everything. The right surface should make the next action obvious.

Starter paths are now mode-specific

0.12 also updates how a new workbench begins.

Starter questions, empty-state guidance, and the main homepage entry now differ depending on whether the user is entering General assistant or Data analysis.

That gives the product a more honest starting point:

• if the task begins with an idea, decision, outline, or draft, General assistant gives a cleaner first step
• if the task begins with a file, table, or governed data source, Data analysis remains the stronger starting point

This is a quiet change, but an important one. A product often teaches users what it is for by how it starts.

Why this release matters

For us, 0.12 is a product-clarity release.

It is not built around one flashy feature. It makes the workbench more accurate about the kind of work the user is actually doing:

• some work begins as conversation
• some work begins as analysis
• many real tasks move from one into the other

The value of Tukun.ai is not that everything turns into chat. The value is that discussion, analysis, evidence, and reusable results can stay on one work path without pretending they are all the same step.

That is what 0.12 makes clearer.