Download the exact captured file
---
name: emem-locate-and-recall
description: Resolves a place name to an emem cell64 address and recalls signed Earth-observation facts there (weather, vegetation indices, elevation, soil, land cover, surface water, forest change). Use when the user asks for a measurable value at a named place ("what is the temperature in Bengaluru", "how high is Denali", "what is the NDVI in the Sundarbans") and wants an answer another agent can verify offline. Returns content-addressed facts, a ready-made emem:fact citation token for each, and an Ed25519 receipt. Reads need no key.
---
# emem-locate-and-recall
> **Network use.** The commands in this skill send and receive JSON (and, where a step says so, an image or a raster file) to `https://emem.dev` only, the service the plugin's MCP server connects to. Nothing they download is executed. This skill ships no scripts. Its commands read no environment variable, key or credential, and the only local files they read are ones an earlier step saved from `https://emem.dev`.
Two calls turn a place name into signed facts: `emem_locate`, then
`emem_recall`, both tools of the plugin's MCP server. Without MCP, send
the same JSON bodies to `POST https://emem.dev/v1/locate` and
`POST https://emem.dev/v1/recall`. Each fact comes back with its own
`emem:fact:` token, so the number never has to travel without its
citation.
## When to invoke
The user asks about a measurable geospatial value at a named place:
- "What's the current 2 m air temperature in Bengaluru?"
- "Show me the NDVI in the Sundarbans."
- "What's the elevation of Mount Kilimanjaro?"
If you already hold a `cell64` (a string like `defi.zb493.zezo.zcb35`),
skip step 1. For an extent rather than a point, use
[`emem-recall-polygon`](../emem-recall-polygon/SKILL.md). For a picture,
`GET https://emem.dev/v1/cells/<cell64>/scene.png`.
## Step 1: resolve the place
Call `emem_locate` with:
```json
{"q": "Bengaluru, India"}
```
Read `cell64`, `place_label`, `via` and `disambiguation_required` in
the result.
`place` is accepted as a synonym of `q`, and `{"lat":..,"lng":..}` skips
the geocoder. When `disambiguation_required` is true, or
`selected.is_high_confidence` is false, read `alternatives` and ask the
user which one they meant rather than citing rank 0. "Springfield"
returns Missouri first and flags it; that is the case this exists for.
## Step 2: recall the bands
Call `emem_recall` with the `cell64` from step 1:
```json
{"cell": "defi.zb493.zezo.zcb35", "bands": ["weather.temperature_2m", "indices.ndvi"]}
```
Save the result's JSON text as `recall.json`. Then print the latest
reading per band, with the token to cite it by:
```sh
jq -r '.current_by_band | to_entries[] | .value' recall.json | while read cid; do
jq -c --arg c "$cid" '.facts[] | select(.fact_cid == $c)
| {band, value_verbatim, unit, observed_at, memory_token}' recall.json
done
```
Read the response this way, not by position:
- `facts[]` holds **every stored reading** for the requested bands, in
`fact_order` (`tslot_ascending`), so `facts[0]` is the oldest, not the
current one.
- `current_by_band` maps each band to the `fact_cid` of its latest
reading. Use it to pick the fact you quote.
- `value_verbatim` is the exact decimal string that was signed. Quote it
rather than re-typing the JSON number.
- `memory_token` on each fact is the `emem:fact:<cell64>:<fact_cid>`
citation. Hand that to the user or the next agent.
- `materialize_notes[]` says what the responder did for each band you
asked for: `materialized`, or `skipped` with a `reason_class`. A
`timeout` note with `retryable: true` means the fetch continues in the
background; the same call a few seconds later usually answers warm.
- `receipt` signs the `fact_cids` this response served. Verify it with
[`emem-verify-receipt`](../emem-verify-receipt/SKILL.md).
Recorded on 2026-09-28 at `defi.zb493.zezo.zcb35` (Bengaluru), in 14 s:
eleven facts came back, nine of them stored `indices.ndvi` readings
from earlier dates and two temperature readings. `current_by_band`
picked `weather.temperature_2m` = `28.0` degC observed
`2026-09-28T10:00:00Z`, token
`emem:fact:defi.zb493.zezo.zcb35:nflpddk7zsncywguwjzk5koksseqfyx4jnngkuryrnd4aykqlpfq`.
The fresh NDVI fetch hit the 14 s materialiser cap and came back as a
`retryable` timeout note, so the NDVI answer was the latest stored one.
The receipt verified offline with `verify.py`.
## Picking bands
Ask the responder, not this page: the catalogue changes per deployment.
- `GET /v1/bands` lists the cube bands and the scalar keys under each
(`.bands[].key`, `.bands[].scalar_keys`).
- `GET /v1/materializers?page_size=100&summary=true` lists the bands this
responder can fetch on a miss (follow `pagination.next_url`).
Common scalars that auto-materialise on emem.dev today:
| Topic | Bands |
|---|---|
| Weather now | `weather.temperature_2m`, `weather.precipitation_mm`, `weather.relative_humidity_2m`, `weather.wind_speed_10m` |
| Climate history | `era5.t2m`, `era5.precip`, `power.t2m`, `terraclimate.precip_normal_mm` |
| Air | `cams.pm25`, `cams.no2`, `cams.o3`, `cams.aod_550` |
| Vegetation | `indices.ndvi`, `indices.evi`, `indices.ndmi`, `modis.ndvi_mean`, `modis.lai_8day` |
| Terrain | `copdem30m.elevation_mean`, `gmrt.topobathy_mean` (ocean depth) |
| Land cover | `esa_worldcover.lc_2021`, `koppen` |
| Surface water | `surface_water.occurrence`, `surface_water.recurrence`, `surface_water.seasonality`, `surface_water.transition_class` |
| Forest | `forest_change.treecover2000`, `forest_change.lossyear` (Hansen GFC), `jrc_gfc2020.forest_2020`, `jrc_tmf.deforestation_year` |
| Soil | `soilgrids.phh2o_0_30cm`, `soilgrids.soc_0_30cm`, `soilgrids.clay_0_30cm` |
The foundation-model embedding bands are retired on emem.dev: facts they
signed before still read and verify, nothing new is computed, and asking
for one returns a note with reason `band_retired_at_this_responder`.
Prefer the deterministic indices, which anyone can recompute from the
same Sentinel scene.
## A value is a pixel sample, not a cell average
`spatial_basis` on every recall says it plainly: the cell (about 9.5 m)
is an address, and the value is the source pixel overlapping it. A
30 m DEM or a 463 m MODIS pixel covers many cells, so never difference
two adjacent cells and call it a gradient.
## Recovering from errors
- `cid_not_found` (404): nothing is stored for that band at that cell and
no materialiser could fetch it. `GET /v1/data_availability` says which
bands can be materialised.
- A malformed `cell64` is a 400 naming the bad symbol. Re-run step 1.
- `compute_timeout`: the request passed the responder's 40 s transport
budget. Ask for fewer bands, or retry once the background fetch lands.
Recall does not take an MCP `task` param: only `emem_eudr_dds` and
`emem_hunt` run as tasks.
- A tool result over the MCP wire budget (about 24 KB) comes back
slimmed, and `_emem_truncation` names what it left out. Ask for
fewer bands, or send the same body to the REST route, which returns
the whole response.
## One-call alternative
`POST /v1/ask {"question":"what is the elevation of Bengaluru?"}` (MCP
`emem_ask`) classifies the question, runs locate and recall for you, and
returns the same signed facts. Use the two-step path above when you need
to control the bands or the place resolution.
## Before you quote the number
Carry the `memory_token`, not the bare number: whoever receives it
resolves the byte-identical signed fact rather than trusting your
summary. See [`emem-referential-drift`](../emem-referential-drift/SKILL.md).
Before the sentence goes out, run
[`emem-verify-before-publish`](../emem-verify-before-publish/SKILL.md).