Provided by Exa
Deep Neural Retrieval by Meaning
Multi-step neural retrieval via Exa's deep mode: the request is expanded into several passes over the page index and the strongest pages are returned, each with title, link, published date and author. Up to five extra query variations, the same category, domain and date filters, and text or highlights per hit. One request per call, up to 10 pages; slower and dearer than the standard mode.
Contract snapshot: 2026-09-17. Inspect in Glasser for current terms.
{
"type": "object",
"$schema": "http://json-schema.org/draft-07/schema#",
"required": [
"query"
],
"properties": {
"query": {
"type": "string",
"maxLength": 500,
"minLength": 1
},
"category": {
"enum": [
"company",
"publication",
"news",
"personal site",
"financial report",
"people"
],
"type": "string"
},
"contents": {
"type": "object",
"properties": {
"text": {
"type": "object",
"properties": {
"maxCharacters": {
"type": "integer",
"maximum": 10000,
"minimum": 1
}
},
"additionalProperties": false
},
"highlights": {
"type": "object",
"properties": {
"query": {
"type": "string",
"maxLength": 500,
"minLength": 1
},
"maxCharacters": {
"type": "integer",
"maximum": 10000,
"minimum": 1
}
},
"additionalProperties": false
}
},
"additionalProperties": false
},
"numResults": {
"type": "integer",
"maximum": 10,
"minimum": 1
},
"excludeDomains": {
"type": "array",
"items": {
"type": "string",
"pattern": "^(?=.*\\.)(\\*\\.)?[a-z0-9-]+(\\.[a-z0-9-]+)*(\\/[^\\s]*)?$",
"maxLength": 253,
"minLength": 4
},
"maxItems": 50,
"minItems": 1
},
"includeDomains": {
"type": "array",
"items": {
"type": "string",
"pattern": "^(?=.*\\.)(\\*\\.)?[a-z0-9-]+(\\.[a-z0-9-]+)*(\\/[^\\s]*)?$",
"maxLength": 253,
"minLength": 4
},
"maxItems": 50,
"minItems": 1
},
"endPublishedDate": {
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}T\\d{2}:\\d{2}:\\d{2}(\\.\\d{1,3})?Z$"
},
"additionalQueries": {
"type": "array",
"items": {
"type": "string",
"maxLength": 300,
"minLength": 1
},
"maxItems": 5,
"minItems": 1
},
"startPublishedDate": {
"type": "string",
"pattern": "^\\d{4}-\\d{2}-\\d{2}T\\d{2}:\\d{2}:\\d{2}(\\.\\d{1,3})?Z$"
}
},
"additionalProperties": false
}{
"query": "startups building solid-state batteries for grid storage",
"numResults": 10,
"additionalQueries": [
"solid-state battery grid storage companies"
]
}This Endpoint publishes no output schema. You get the Provider's own response body, unchanged — we do not warrant a shape we cannot guarantee.
How to run it
Two fields identify it: the Provider and the endpoint key. The key looks like a path because it usually mirrors the Provider's own, but it is an opaque identifier — not a URL you can open. The same provider:endpoint pair is what your Policy lists.
Once: paste this into your agent's chat. The Skill installs the CLI and logs in by itself. For an agent that cannot use a shell, connect it over MCP instead.
set up https://glasser.ai/SKILL.mdThen ask for this Endpoint:
Run the Exa Endpoint /search?type=deep on Glasser with input {"query":"startups building solid-state batteries for grid storage","numResults":10,"additionalQueries":["solid-state battery grid storage companies"]}glasser run -p exa -e '/search?type=deep' -i '{"query":"startups building solid-state batteries for grid storage","numResults":10,"additionalQueries":["solid-state battery grid storage companies"]}' --endpoint-version 1Pinned to version 1, the one this page shows: if a newer version is published first, the Run is refused instead of charged under new terms. Install and log in first on Get Started.
Once per Run:
IDEMPOTENCY_KEY=$(uuidgen)Run it, and copy this same block to send it again:
curl -X POST https://api.glasser.ai/v1/runs \
-H "Authorization: Bearer gl_..." \
-H "Idempotency-Key: $IDEMPOTENCY_KEY" \
-H "Content-Type: application/json" \
-d '{"provider":"exa","endpoint":"/search?type=deep","input":{"query":"startups building solid-state batteries for grid storage","numResults":10,"additionalQueries":["solid-state battery grid storage companies"]},"endpoint_version":1}'Reuse the same Idempotency-Key to send it again: the second send is a read of the first Run. A new key is a new Run and a new Charge. Pinned to version 1 for the same reason as the CLI.
What each outcome charges
| No result | The Provider answered: nothing matched. | $0.00 |
|---|---|---|
| Provider error | The Provider failed to answer. | $0.00 |
| Timed out | The deadline won. | $0.00 |
| Internal failure | Our defect. | $0.00 |