The kong setup command is an interactive wizard that configures your LLM providers and verifies your environment. You only need to run it once, but you can re-run it anytime to change your configuration.
Step 1: Choose Your Providers
The wizard starts by asking which LLM providers you want to use:
Step 1: Which LLM providers would you like to use?
1) Anthropic (Claude)
2) OpenAI (GPT-4o)
3) Custom endpoint (OpenAI-compatible)
4) Anthropic + OpenAI
| Provider | Default Model | Setup |
|---|
| Anthropic | claude-opus-4-6 | Set ANTHROPIC_API_KEY env var |
| OpenAI | gpt-4o | Set OPENAI_API_KEY env var |
| Custom | User-specified | Configure via kong setup or --base-url flag |
Step 2: API Key Verification
For standard providers (Anthropic and OpenAI), Kong checks whether the corresponding environment variable is set:
ANTHROPIC_API_KEY for Anthropic (Claude)OPENAI_API_KEY for OpenAI (GPT-4o)
If a key is found, the wizard displays a masked preview (e.g., sk-ant-...abcd). If a key is missing, it shows you exactly where to get one and the export command to set it.
Kong reads API keys from environment variables, not from the config database. Set them in your shell profile (~/.zshrc or ~/.bashrc) so they persist across sessions. The wizard verifies they are present but does not store them.
Custom Endpoint Configuration
If you choose option 3 (Custom endpoint), the wizard walks you through configuring an OpenAI-compatible endpoint. This works with local inference servers and third-party providers:
| Field | Description | Examples |
|---|
| Endpoint URL | Base URL of the OpenAI-compatible API | http://localhost:11434/v1 (Ollama), http://localhost:8000/v1 (vLLM), https://openrouter.ai/api/v1 (OpenRouter) |
| Model name | The model identifier to use | llama3.1, deepseek-coder, anthropic/claude-3.5-sonnet |
| API key | Authentication key (leave blank for local servers) | — |
| Max prompt size | Maximum prompt size in characters | Default from built-in limits |
| Max functions per batch | Maximum functions analyzed per LLM call | Default from built-in limits |
| Max output tokens | Maximum tokens in LLM response | Default from built-in limits |
Default Provider Selection
If you enable multiple providers (option 4 — Anthropic + OpenAI), the wizard asks which one should be the default:
Step 3: Which provider should be the default?
1) Anthropic (Claude)
2) OpenAI (GPT-4o)
kong analyze without a --provider flag. You can always override it per-analysis:
# Uses your default provider
kong analyze ./binary
# Override for this run
kong analyze ./binary --provider openai
Ghidra Detection
After provider configuration, the wizard checks for a Ghidra installation. Kong auto-detects Ghidra in common locations. If found, it displays the path. If not, it provides installation instructions:
Ghidra
Found: /opt/homebrew/Caskroom/ghidra/11.3.1/ghidra_11.3.1_PUBLIC
GHIDRA_INSTALL_DIR environment variable:
export GHIDRA_INSTALL_DIR="/path/to/ghidra"
What Gets Saved
The setup wizard saves your configuration to a SQLite database at:
This location can be overridden with the KONG_CONFIG_DIR environment variable.
The database stores:
- Enabled providers — which providers are available for analysis
- Default provider — which provider to use when no
--provider flag is given
- Custom endpoint config — base URL, model name, API key, and limit overrides (if configured)
- Setup completion flag — so Kong knows the wizard has been run
API keys are not stored in the config database. They are always read from environment variables at runtime. This keeps secrets out of the config file.
Re-Running Setup
You can re-run kong setup at any time to change providers, switch defaults, or update custom endpoint settings. The wizard shows your current configuration and lets you reconfigure everything.
Next Steps
