The name may be wrong
Misspellings, phonetic fragments, family nicknames, and borrowed-language names are normal inputs — not failures.
Someone remembers a dish they haven't eaten in years. A smell, a nickname, a texture — not a recipe. Achiote turns those fragments into the one bite that proves you found it.
Type something half-remembered and our AI will walk you through the full reconstruction pipeline.
Open the food memory detectiveA person may only remember a smell, a texture, a nickname, or the person who made it. Achiote organizes those fragments into research instead of guessing.
Misspellings, phonetic fragments, family nicknames, and borrowed-language names are normal inputs — not failures.
The thing that matters may be one browned edge, one spice bloom, one sour note, one chewy bite, or one table ritual.
What someone remembers, what research found, what the AI guessed, and what nobody knows — kept separate so nothing pretends to be certain.
Your AI does the searching. Achiote keeps the work honest and organized.
Capture rough names, region clues, ingredients, sensory details, and the gaps in what someone remembers — then ask gentle follow-up questions.
Turn clues into hypotheses, facts to verify, and the right questions to ask family members before committing to a dish.
What someone said, what research found, what the AI guessed, and what's still unknown — all labeled, all visible.
A single bite, sip, or aroma using ingredients from any grocery store — before anyone shops for something obscure or commits to a full recipe.
Once a cue works, suggest where to find the real ingredients near where they live now — and what to substitute if something isn't available.
{
"title": "Minimum viable composed-bite cue",
"format": "bite",
"effortMinutes": 15,
"accessibilityPrinciples": [
"use grocery-store carriers and proteins first",
"test in one or two bites",
"use pantry aromatics before specialty sourcing"
],
"substituteLogic": [
"Proteins and fats carry Maillard notes and fat-soluble aromatics.",
"Starches control texture and sauce absorption.",
"Acid, sugar, and salt move the bite toward the remembered balance."
],
"followUpIfItWorks": [
"Ask which part hit first: smell, texture, sauce, fat, spice, or sweetness.",
"Use source_ingredients to help the user find items near where they live."
]
}
Each one follows the same path: gather what someone remembers, research, find the sensory trigger, test it with one bite, then help them find ingredients nearby.
"My mom said my abuela made something that sounded like pass-teh-lay. Maybe plantains or pork?"
"A soup my aunt made. Sour, pale, lots of dill, maybe dairy, but nobody remembers the name."
"I ate shark one time in Trinidad. Fried, in some kind of bread, with a green sauce."
Achiote runs as a stdio MCP server. Build it, point your agent at the output, and it's ready.
# Install dependencies and build npm ci npm run build # Claude Code (.mcp.json) { "mcpServers": { "achiote": { "command": "node", "args": ["/path/to/achiote/dist/index.js"] } } } # Codex (~/.codex/config.toml) [mcp_servers.achiote] command = "node" args = ["/path/to/achiote/dist/index.js"] enabled = true startup_timeout_sec = 10
Achiote makes your AI more useful by making uncertainty visible and the next test smaller.