One workflow, four surfaces
The same reconstruction runs from the MCP server, CLI, API, and the web app.
How it works
Research first. Then the recipe.
Achiote reconstructs a half-remembered dish. It takes your memory clues, plans research, reasons over food science, and returns a Memory Receipt, with local sourcing, language clues, and evidence boundaries intact. Use it as an MCP server, a CLI, or an API.
The workflow
From a memory to a receipt
Each step does one job. Together they investigate before they answer, and keep the evidence honest.
Intake
Your memory clues become data, not instructions.
Plan
A source-first path, with confirm/reject signals.
Research
Names, regions, techniques, and competing versions.
Reason
Food science connects flavor, culture, and access.
Receipt
A clear result with the unknowns kept visible.
What you get
A result you can trust
Real research. Real sources.
Outputs separate user-said, researched, inferred, and unknown evidence.
A recipe that remembers
Local sourcing and food-science reasoning come before any full dish.
Example Memory Receipt
DishSancocho de gallina
OriginDominican Republic
Memory notesSunday gatherings, cilantro, yuca, plantains, comfort.
First tasteBuild the broth aroma first, then test the starch texture.
What every receipt includes
A human-readable result, with the workflow and the evidence trail kept transparent.
- Sensory analysis and mechanism reasoning
- Ingredient mapping with accessible substitutes
- Local sourcing before specialty exactness
- Language clues preserved as evidence
- One minimum-viable taste before the full roadmap
The recipe was never written down. The memory was. And research can bring it back to the table.