Skip to content
KASECODE
app.header.homeapp.header.docs

Kase Features

What makes KaseCode different from the base it started from.

KaseCode is no longer just a rename. It ships with a set of workflow and UX changes tuned for faster day-to-day use.

Latest production layer

KaseCode now has routing memory, provider checks, and shared app/TUI controls.

Use this page as the fast map for what changed beyond the base project: Auto routing, Caveman mode, /btw, /close, NVIDIA NIM, first-run desktop setup, and purple-first UI polish.

Command

kase

Default style

Caveman on

Auto mode

Static fast path by default

Advisor

OpenRouter free can advise only

Built for fast terminal work

KaseCode keeps the terminal-first workflow, but adds opinionated changes that make it feel lighter and more interruptible while you work.

  • kase is the main CLI command.
  • The TUI ships with Kase branding and a purple-first visual theme.
  • The desktop app and TUI both expose shared Auto model controls.
  • Thinking output can stay hidden while still letting the model reason in the background.
  • Sessions compact automatically before context gets too close to full.
  • Desktop settings include provider tests and first-run setup shortcuts.
  • Auto decisions are persisted in message metadata so you can audit routing later.

Kase-only slash commands

/btw and /aside

Use /btw (or its alias /aside) to ask a side question or add an in-between instruction while a conversation or ongoing task is already in motion.

/btw also make the sidebar tighter and keep the previous work
/aside did you include the newest imports?

This is useful when you do not want to fully redirect the current thread but still want to inject a correction, reminder, or extra ask.

Generic Slash Commands

The prompt bar in both the desktop app and TUI now supports triggering ANY built-in command directly by typing its name with a leading slash. This lets you quickly run workflow commands without opening the command palette:

  • /undo: Undo the last message
  • /redo: Redo an undone message
  • /compact: Force a session compaction
  • /fork: Fork the current session
  • /share: Open the share/export dialog
  • /clear: Clear the current prompt

Any command registered in the KaseCode command system can be triggered this way.

/close

/close is a direct alias for exiting the TUI.

/close

It behaves the same as /exit, /quit, and /q.

/caveman

KaseCode includes built-in Caveman mode for compressed answers.

/caveman on
/caveman lite
/caveman full
/caveman ultra
/caveman off

Caveman mode is enabled by default and can be changed at any time during a session.

/auto

Use /auto to tune how KaseCode picks a model when you keep the model selector on Auto.

/auto quality
/auto reasoning on
/auto prefer provider openai anthropic
/auto avoid model mini nano
/auto advisor on

This updates shared Auto rules used by both the TUI and desktop app.

/auto-rules

Use /auto-rules in the TUI to open the dedicated Auto Rules dialog.

/auto-rules

The dialog lets you change:

  • mode
  • reasoning requirement
  • preferred providers
  • avoided providers
  • preferred model patterns
  • avoided model patterns
  • OpenRouter advisor mode

The sidebar now exposes extra working-state information so you can glance at what is happening without opening more UI:

  • context usage
  • current model
  • provider
  • Caveman mode status

This makes it easier to stay oriented in long sessions.

Shared Auto model routing

KaseCode includes an Auto model option in both the desktop app and TUI.

When Auto is selected, KaseCode can route each prompt to the best connected model based on:

  • quality vs speed vs cost mode
  • reasoning requirement
  • preferred providers
  • avoided providers
  • preferred model name patterns
  • avoided model name patterns
  • media requirements such as image, PDF, audio, or video support
  • optional advisor mode

Both app surfaces now expose dedicated UI for this:

  • Desktop app: Settings -> Models -> Auto Rules
  • TUI: model picker -> Auto Rules... or /auto-rules

The prompt bar also shows why Auto picked the current model, so routing is not a black box. After a message is sent, KaseCode also keeps the Auto reason in the message metadata, so you can audit which rule or advisor choice routed the request.

If you also connect OpenRouter, KaseCode can use openrouter/free as a lightweight advisor for model choice. That advisor helps pick from your configured models, but it is not the model that performs the real task. Turn it on from either surface:

/auto advisor on
/auto advisor off

In the desktop app, open Settings -> Models -> Auto Rules and toggle OpenRouter advisor.

openrouter/free is excluded from execution ranking and fallback execution, so it cannot complete real work even if every other model is unavailable.

Desktop app production controls

The desktop app now has first-run shortcuts and provider/model management built into the main flow:

  • Add provider opens the provider connection panel.
  • Tune Auto opens model settings with Auto Rules.
  • Open project starts a workspace.
  • Quick commands points users at /btw, /caveman, /auto, and /close.

Provider settings include a Test action for connected providers. It refreshes provider metadata and checks that the provider is connected and returning usable models.

Auto Rules in desktop settings include:

  • mode
  • reasoning requirement
  • OpenRouter advisor toggle
  • preferred providers
  • avoided providers
  • preferred model patterns
  • avoided model patterns
  • live preview of the model Auto would pick

This makes the desktop app suitable for setting up providers without memorizing config keys.

Automatic compaction

KaseCode can compact sessions before the context window becomes a hard problem.

By default, auto-compaction starts around 85% of usable context instead of waiting until the session is almost exhausted.

You can tune this with config:

kasecode.json
{
  "compaction": {
    "auto": true,
    "threshold": 0.85,
    "reserved": 20000
  }
}

NVIDIA NIM support

KaseCode supports NVIDIA NIM through the OpenAI-compatible API surface.

Use the NVIDIA-hosted endpoint:

https://integrate.api.nvidia.com/v1

Then connect it from the TUI:

/connect

After saving your API key, NIM models can appear in the model picker and be used like other providers.

See Providers for setup details.

Branding and release surface

KaseCode also replaces the old upstream-facing product surface with Kase-specific defaults:

  • website and docs point to kase.ammar.click
  • repo and release references point to kyssta-exe/KaseCode
  • legal and product copy use Kyssta
  • desktop and CLI product names use KaseCode

What is still intentionally external

Some references may still use the old opencode-* package names when they refer to third-party packages that have not been forked yet.

Those are ecosystem dependency names, not the shipped KaseCode product name.