tk_ helper

Use your own AI subscriptions through Toolkit-LLM — one CLI, two superpowers.

npx @toolkit-cli/tk-helper

Why tk-helper?

You're paying for AI twice. You have a ChatGPT subscription AND you're paying per-token for Claude Code. tk-helper fixes both:

| Without tk-helper | With tk-helper | |---|---| | Claude Code → Anthropic API ($$$) | Claude Code → Toolkit self-hosted models (your API key) | | ChatGPT subscription sits unused in browser | ChatGPT subscription powers Toolkit's web app | | Two separate AI products | One unified system |

Quick Start

npx @toolkit-cli/tk-helper

First run opens the setup wizard:

  ╔════════════════════════════════════════╗
  ║                                        ║
  ║      tk_ helper                        ║
  ║      v0.2.0                            ║
  ║                                        ║
  ║  Connect Claude Code & ChatGPT to      ║
  ║  Toolkit-LLM's self-hosted models      ║
  ║                                        ║
  ╚════════════════════════════════════════╝

  Step 1: Toolkit-LLM API Key
  Get your key at https://toolkit-llm.com/dashboard/keys

  API Key (tk_live_...): ████████
  Validating... ✓ Valid

  Step 2: Configure Claude Code
  Map Claude Code → Toolkit-LLM? (Y/n): Y
  ✓ Claude Code configured

  Step 3: Connect ChatGPT (optional)
  Connect ChatGPT? (y/N): y
  Opening browser for OpenAI authentication...
  ✓ ChatGPT connected

  ✓ Setup complete!

Use with Claude Code

Route Claude Code through Toolkit-LLM's self-hosted models. Same Claude Code UX, different backend.

Setup

npx @toolkit-cli/tk-helper init

Enter your API key. Say yes to "Configure Claude Code". Done.

What happens

tk-helper writes ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "tk_live_YOUR_KEY",
    "ANTHROPIC_BASE_URL": "https://api.toolkit-llm.com/v1",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "toolkit-code-backend",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "toolkit-base",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "toolkit-code-fast"
  }
}

Model mapping

| When Claude Code asks for | Toolkit sends it to | What it is | |---|---|---| | Opus (heavy tasks) | toolkit-code-backend | 27B dense, architecture & APIs | | Sonnet (default) | toolkit-base | 30B GLM, business & analysis | | Haiku (fast tasks) | toolkit-code-fast | 9B, quick edits & autocomplete |

Use it

claude "build me a landing page"

That's it. Claude Code works exactly the same — it just uses Toolkit's models.

Change model profile

npx @toolkit-cli/tk-helper
# Select "4. Select Model" from the menu

Undo it

npx @toolkit-cli/tk-helper uninstall

Removes only Toolkit's env vars from Claude Code. Everything else untouched.

Use with ChatGPT Subscription

Already paying $20/mo for ChatGPT Plus? Use that subscription through Toolkit's web app.

Connect

npx @toolkit-cli/tk-helper connect

Opens your browser → sign into ChatGPT → authorize → done.

Start the bridge

npx @toolkit-cli/tk-helper start

Runs a local server on localhost:1455. The web app at toolkit-llm.com auto-detects it.

How it works

toolkit-llm.com → detects localhost:1455 → routes through your ChatGPT token → GPT-5.4

You get GPT-5.4 in Toolkit's workspace at your flat subscription rate. No per-token billing.

Disconnect

npx @toolkit-cli/tk-helper disconnect

All Commands

tk-helper              Interactive menu (wizard on first run)
tk-helper init          Re-run setup wizard
tk-helper start         Start bridge server (localhost:1455)
tk-helper connect       Authenticate with ChatGPT
tk-helper disconnect    Clear ChatGPT tokens
tk-helper claude-code   Configure Claude Code → Toolkit-LLM
tk-helper uninstall     Remove from Claude Code + clear tokens
tk-helper doctor        Health check
tk-helper status        Show current config
tk-helper help          Show help

Health Check

npx @toolkit-cli/tk-helper doctor

  Doctor — Health Check

  1. API Key............. ✓ Valid
  2. Claude Code......... ✓ Installed
  3. Claude Settings..... ✓ Toolkit configured
  4. ChatGPT OAuth....... ✓ Connected
  5. Bridge Server....... ○ Not running

  Result: 4 passed, 0 issues

Files

| Path | What | Permissions | |---|---|---| | ~/.tk-helper/config.json | API key + model | 0600 | | ~/.tk-helper/tokens.json | ChatGPT OAuth tokens | 0600 | | ~/.claude/settings.json | Claude Code env vars | Default |

Full cleanup

npx @toolkit-cli/tk-helper uninstall
rm -rf ~/.tk-helper

Models

| Name | ID | Best for | Context | |---|---|---|---| | Default | toolkit-chat | Fast everyday chat | 32K | | Pro Chat | toolkit-chat-pro | Deeper reasoning | 32K | | Assistant | toolkit-base | Business & analysis | 200K | | Builder | toolkit-code | Full-stack coding | 131K | | Backend | toolkit-code-backend | Architecture & APIs | 65K | | Voice | toolkit-voice | Real-time speech | 512 | | Camera | toolkit-cam | Visual understanding | 1K |

API

Toolkit-LLM is OpenAI-compatible. Works with any SDK:

from openai import OpenAI

client = OpenAI(
    api_key="tk_live_YOUR_KEY",
    base_url="https://api.toolkit-llm.com/v1"
)

response = client.chat.completions.create(
    model="toolkit-code",
    messages=[{"role": "user", "content": "Hello"}]
)

curl https://api.toolkit-llm.com/v1/chat/completions \
  -H "Authorization: Bearer tk_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"toolkit-code","messages":[{"role":"user","content":"Hello"}],"stream":true}'

toolkit-llm.com · Quickstart · Pricing · Models · MIT License