oktk - CLI Output Compressor for LLMs

The Problem

When you run commands through an AI assistant, the full output goes into the LLM context:

$ git status
# Returns 60+ lines, ~800 tokens
# Your AI reads ALL of it, you pay for ALL of it

Every token costs money. Verbose outputs waste your context window.

The Solution

oktk sits between your commands and the LLM, compressing outputs intelligently:

┌──────────┐     ┌──────────┐     ┌──────────┐
│ Command  │ ──► │   oktk   │ ──► │   LLM    │
│ (800 tk) │     │ compress │     │ (80 tk)  │
└──────────┘     └──────────┘     └──────────┘
                      │
                 90% SAVED

When Does It Work?

Automatically when you run supported commands through OpenClaw:

Concrete Example

Before oktk (800 tokens sent to LLM):

On branch main
Your branch is ahead of 'origin/main' by 3 commits.
  (use "git push" to publish your local commits)

Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git restore <file>..." to discard changes in working directory)
        modified:   src/components/Button.jsx
        modified:   src/components/Header.jsx
        modified:   src/utils/format.js
        modified:   src/utils/validate.js
        modified:   package.json
        modified:   package-lock.json

Untracked files:
  (use "git add <file>..." to include in what will be committed)
        src/components/Footer.jsx
        src/components/Sidebar.jsx
        tests/Button.test.js

no changes added to commit (use "git add" and/or "git commit -a")

After oktk (80 tokens sent to LLM):

📍 main
↑ Ahead 3 commits
✏️  Modified: 6
❓ Untracked: 3

Same information. 90% fewer tokens. Same cost savings.

How It Works Technically

1. Intercepts command output after execution
2. Detects command type (git? npm? ls?)
3. Applies specialized filter for that command
4. Extracts only essential information
5. Caches results (same command = instant, no reprocessing)

Safety First

oktk never breaks your workflow: