Kaimei Labs

Deterministic verification infrastructure for AI agent outputs. Your LLM authors, Guardian judges. Recipes are the first vertical — the same approach generalises to any procedural domain where correctness matters.

Guardian Engine

When AI agents generate recipes, they hallucinate — impossible temperatures, skipped techniques, wrong ingredients, broken emulsions. Guardian Engine is a deterministic oracle inside your agent’s generate→verify loop: it catches these errors before they reach the pan and returns machine-actionable patches so the agent can fix exactly what’s wrong.

An LLM critique is a sample — it misses differently on every run. Guardian’s symbolic engine makes guarantees a generative model structurally cannot:

Exhaustive — every rule is checked on every call, not a sample of them.
Certified negatives — “no EU Annex II allergen source detected” is an absence claim an LLM cannot make.
Replayable — same input + same spec + same knowledge-base version → byte-identical verdict, pinned by kb_version_hash and master_hash for audit.
Machine-actionable repair — structured patches, plus a deterministic fix_recipe tool (no LLM in the loop).
Bring your own spec — verify against your house recipe or SOP via master_json, not just our catalog.

161 master recipes across 5 regions — from Confit de Canard to Bulgogi to Jerk Chicken — with new dishes added regularly. Seven MCP tools: verify_recipe, fix_recipe, list_dishes, get_master, check_safety, check_allergens, and verify_dietary_claim (vegan / vegetarian / gluten-free / dairy-free / nut-free / halal / kosher).

Connect in 30 Seconds

Guardian is a hosted MCP server. No install, no API key, no Docker.

Claude Desktop — add to claude_desktop_config.json:

{
  "mcpServers": {
    "guardian": {
      "url": "https://api.kaimeilabs.dev/mcp",
      "transport": "streamable-http"
    }
  }
}

Cursor — add via Settings → MCP Servers:

{
  "guardian": {
    "url": "https://api.kaimeilabs.dev/mcp",
    "transport": "streamable-http"
  }
}

VS Code — add to .vscode/mcp.json:

{
  "servers": {
    "guardian": {
      "type": "http",
      "url": "https://api.kaimeilabs.dev/mcp"
    }
  }
}

Python SDK — pip install mcp httpx:

from mcp.client.session import ClientSession
from mcp.client.streamable_http import streamable_http_client
result = await session.call_tool("verify_recipe", arguments={"dish": "carbonara", "candidate_json": recipe_json})

What Does a Verification Report Look Like?

Here’s the (abridged) response structure when Guardian catches issues in an AI-generated recipe:

{
  "response_format_version": "v3",
  "kb_version_hash": "7dda40a3b646",
  "verdict": "FAILED",
  "matched_against": "Pasta alla Carbonara (Master)",
  "master_source": "catalog",
  "master_hash": "3357e09...41fd29",
  "summary": {"CRITICAL": 4, "WARNING": 0, "INFO": 3},
  "findings": [
    {
      "issue": "TEMPERATURE_MISMATCH",
      "severity": "critical",
      "justification": "Temperature is significantly outside the required range.",
      "details": {"expected": "100.0-130.0", "observed": "180"}
    },
    {
      "issue": "INGREDIENT_SUBSTITUTED",
      "severity": "critical",
      "justification": "'bacon' is in the same group ('cured_pork') as 'guanciale' but is not the canonical ingredient for this recipe.",
      "details": {"expected": "guanciale", "observed": "bacon"}
    }
  ],
  "allergens": [
    "Allergen detected: Milk and products thereof (including lactose)",
    "Allergen detected: Cereals containing gluten (wheat, rye, barley, oats, spelt, kamut)"
  ],
  "patches": [
    {"action": "set_temperature", "step_index": 2, "value": "100.0-130.0"},
    {"action": "replace_ingredient", "remove": "bacon", "add": "guanciale"}
  ]
}

Each finding includes a severity, a justification grounded in culinary science, and a machine-actionable patch — so the agent fixes only what’s wrong instead of guessing. The kb_version_hash and master_hash pin the exact knowledge-base and spec versions, making every verdict replayable as an audit record.

⚠️ Safety note: Guardian output — including allergen warnings and dietary-claim checks — is automated and informational only. It is not food-safety, medical, or certification advice, and a PASSED verdict is never a guarantee that a food is allergen-free or safe for any individual. See the Terms of Service.

Resources

API Docs & SDK — Full schema, Python examples, integration tests
Smithery Listing — One-click install for Claude & Cursor
Terms of Service — Free early access, fair use, data policy
API Endpoint: https://api.kaimeilabs.dev/mcp (Streamable HTTP / MCP)

Why Pass the Prompt?

When you include the user’s original request via original_prompt, Guardian personalises findings to the user’s stated dietary needs and flavour preferences, and activates audience-sensitive safety checks (e.g. flagging honey in recipes for infants, raw egg for pregnant users). Without it, Guardian still returns the full verdict and all findings.

Building an AI cooking assistant, smart kitchen platform, or agentic food-tech product? Contact: partners@kaimeilabs.dev