Self-host CodeWiki on private code: validate the config before you spend tokens

The Stack

Steps

1. 1

   Scaffold the config and validate provider + models (no key)

   Set the provider and both models (CodeWiki uses a main model and a cluster model), and keep the API key out of the config, it belongs in the keychain. CI checks the config parses and the fields are set with no leaked key. Running codewiki generate on your code is fenced.

   cat > codewiki-config.json <<'JSON'
   {
     "provider": "openai-compatible",
     "main_model": "claude-sonnet-5",
     "cluster_model": "claude-sonnet-5",
     "subscription_mode": "claude",
     "api_key_env": "ANTHROPIC_API_KEY"
   }
   JSON
   node -e '
   const fs = require("fs");
   const c = JSON.parse(fs.readFileSync("codewiki-config.json", "utf8"));
   function bad(m) { console.error("BAD: " + m); process.exit(1); }
   if (!c.provider) bad("no provider set");
   if (!c.main_model) bad("main_model must be set (the model that writes the docs)");
   if (!c.cluster_model) bad("cluster_model must be set (the model that clusters the codebase)");
   if (c.api_key && String(c.api_key).length > 0) bad("do not put a raw api key in the config; keys belong in the system keychain (reference by env name)");
   console.log("config OK: CodeWiki provider " + c.provider + ", main-model " + c.main_model + ", cluster-model " + c.cluster_model + "; no key in config (keychain); generation fenced");
   '

2. 2

   Generate the docs on your code (the model step, not checked by CI)

   pip install the CLI, run codewiki config validate to confirm your live config, then codewiki generate --github-pages --create-branch (or --update to regenerate only what changed). This makes real LLM calls on your code; the doc quality and cost are fenced.