diff --git a/format-ingestion.md b/format-ingestion.md
new file mode 100644
index 00000000..f3ae3613
--- /dev/null
+++ b/format-ingestion.md
@@ -0,0 +1,121 @@
+# Format of data ingested in the database
+
+The management script is able to ingest data represented in a JSON format.
+Different files are necessary:
+
+* `neurons.json` that encodes the information about the neurons in general
+* `datasets.json` that encodes the information about the different datasets
+* `connections/xxx.json` that encodes the different connections for dedicated datasets
+* `annotations/xxx.json` that encodes annotatinos for different zones of the anatomy
+
+Those files are automatically exported from third-party tool and shouldn't be edited manually.
+
+## Format of `neurons.json`
+
+This file defines a list of JSON object as root structure:
+
+```json
+[
+    {
+        ...  // definition of neuron 1
+    },
+    ...,
+    {
+        ...  // definition of neuron N
+    }
+]
+```
+
+Each JSON object represents a neuron with this schema:
+
+```json
+{
+    "inhead": int,      // int used as bool, is the neuron part of the head or not
+    "name": string,     // name of the neuron, can be same as classes, or L or R of classes
+    "emb": int,         // int used as bool
+    "nt": string,       // neurotransmitter type
+    "intail": int,      // int used as bool
+    "classes": string,  // general name of the neuron
+    "typ": string       // type of the neuron: "i" =>  ,  TODO fillme
+}
+```
+
+
+## Format of `datasets.json`
+
+This file defines a list of JSON object as root structure.
+
+```json
+[
+    {
+        ...  // definition of dataset 1
+    },
+    ...,
+    {
+        ...  // definition of dataset N
+    }
+]
+```
+
+Each JSON object represents a specific dataset with this schema:
+
+```json
+{
+    "id": string           // unique ID for the dataset
+    "name": string         // display name of the dataset
+    "type": string         // type of dataset: "complete" or "head"
+    "time": int            // time of the dataset
+    "visualTime": int      // visualTime of the dataset
+    "description": string  // description of the dataset
+    "axes": [              // OPTIONAL: different axes and their representation, not used but can appear in the file
+        ...
+    ]
+}
+```
+
+## Format of `connections/xxx.json`
+
+The `connections` directory encodes the information about the different connections by dataset.
+Each file in this directory is named after the `id` of a dataset present in the `datasets.json` file, e.g.: a dataset defined using the `id` `white_1986_jsh` will defines each of the connections of the dataset in the file `connections/white_1986_jsh.json`.
+
+Each of those files is a list of JSON object where each of the JSON objects encodes different connections between different neurons.
+The schema is the following:
+
+```json
+{
+    "ids": [ ... ],       // a list of int, where each int represents the ID of the neurons involved in this connection
+    "post": string,       // the name of a neuron as defined in "neurons.json"
+    "post_tid": [ ... ],  // a list of int where each int represents the ID of a post synapse for a dedicated post neuron
+    "pre": string,        // the name of a neuron as defined in "neurons.json"
+    "pre_tid": [ ... ],   // a list of int where each int represents the ID of a pre synapse for a dedicated pre neuron
+    "syn": [ ... ],       // a list of int where each int represents the weight of a post or pre synapses (indice matches the neuron in pre/post_tid)
+    "typ": int            // the type of connection ("electrical" or "chemical")
+}
+```
+
+For each of those objects: `ids`, `post_tid`, `pre_tid` and `syn` need to have the same number of elements.
+
+## Format of `annotations/xxx.json`
+
+The `annotations` directory encodes annotations about the different part (`head` or `complete`) following the naming convention `part.annotations.json`, e.g.: the annotations for the `head` are located in `annotations/head.annotations.json`.
+
+Each of those files is a JSON object that defines categories as keys and a list of neurons couples as values.
+Here is the schema for the `head.annotations.json` file (the `complete.annotations.json` file, while existing, is an empty JSON object).
+
+```json
+{
+    "increase": [    // the type of annotation
+        [
+            string,  // pre, the ID/name of a neuron from "neurons.json"
+            string   // post, the ID/name of the other neuron from "neurons.json" that is part of the couple
+        ]
+    ]
+}
+```
+
+The types of annotations can be `increase`, `variable`, `postembryonic`, `decrease` or `stable`
+
+### Note:
+
+The existing repository contains a `trajectories` folder with a set of JSON files.
+Those files are not ingested anymore, they are part of a legacy system.