SourceLace Docs
FAQOpen the app

Search index: OpenSearch, Elasticsearch, Azure AI Search

Your own search index, searched from SourceLace. Many organizations already index their files (or other content) in a search service they run in their own cloud. Connected as a Search index source, that index answers searches from the SourceLace app, AI apps and agents, with a link back to each file. SourceLace copies nothing: each search goes to your index when someone asks, and the hits are held only like any other result (up to 30 minutes).

Status: Pilot. New or partial, offered to pilot customers. It is built on each service's published search API and has not been used with a customer's live index yet.

At a glance

Connects through Sign-in Network Status First test question
Search index (OpenSearch, Elasticsearch, Azure AI Search) The service's search API over HTTPS Personal sign-in with oauth; shared account with an API key or login Allow SourceLace's IP addresses, if the service limits them Pilot Test connection on Connections, then "Find the Corvanta renewal in search_index:docs."

Add each source with the usual steps. The network answers are explained in What your network needs.

When to use it

  • Your files are in Amazon S3, Azure Storage or Google Cloud Storage and you already index them. Your index searches everything it holds at once, so it suits large stores and files people search all the time. For a few folders you do not index, an S3, Azure Storage or Google Cloud Storage source can search inside files itself: see How S3, Azure Storage and Google Cloud Storage are searched.
  • One index, one source. Each source searches the one index you name. Add a source for each index people should search.
  • Searching only. A Search index source finds things; it does not open or change files. Each hit links back to the file where it is kept, and if that store is also a SourceLace source, people can read the file there.

How people sign in

Kind: Search index (search_index). The admin picks how people sign in (auth):

Choice What it means Works with
Each person's own account (oauth, recommended) Each person signs in as themselves, so your search service applies its own permissions to each search, including document-level security where you set it up. Azure AI Search with Microsoft Entra ID. Elasticsearch and OpenSearch when your cluster accepts sign-in through your identity provider.
A shared API key (api_key) Everyone searches as one key. Labelled Shared account everywhere; use access by group to choose who may use the source. Elasticsearch (an API key) and Azure AI Search (a query key).
A shared login (basic) Everyone searches as one user name and password. Labelled Shared account. Elasticsearch and OpenSearch.

With a shared account, create a key or user that can only read the one index, nothing else.

1. Prepare the index (your search service's admin)

Follow the part for your search service. About 20 minutes.

Who: an Azure Owner or User Access Administrator on the search service, and, for each person's own sign-in, a Microsoft Entra Application Administrator.

Each person's own account (oauth, recommended):

  1. In the Azure portal, open the search service. Under Settings → Keys, set API access control to Role-based access control or Both, and save. With the default (API keys only), Azure refuses everyone's sign-in.
  2. Under Access control (IAM) → Add → Add role assignment, choose Search Index Data Reader and add the people or groups who should search through SourceLace.
  3. In the Microsoft Entra admin center → App registrations → New registration, name it "SourceLace search", keep Accounts in this organizational directory only, and add the redirect URI type Web with exactly https://mcp.sourcelace.ai/connect/callback.
  4. In the new app, API permissions → Add a permission → APIs my organization uses, choose Azure AI Search (also listed as Azure Cognitive Search), then Delegated permissions → user_impersonation. Optionally Grant admin consent, so people are not asked one by one.
  5. Certificates & secrets → New client secret. Copy its Value at once (it is shown only once). The Secret ID next to it is not the secret.
  6. From the app's Overview, copy the Application (client) ID and the Directory (tenant) ID.
  7. Give the SourceLace admin: the service's address (https://<service>.search.windows.net, from the service's Overview), the index name, the client ID, the secret and the tenant ID.

A shared query key (api_key): under Settings → Keys → Manage query keys → Add, name it "SourceLace" and copy the key. A query key can only search.

Links: the Azure AI Search index made by the Import data wizard keeps each file's address in metadata_storage_path, encoded; SourceLace decodes it into a link. Microsoft's own guide: Azure AI Search documentation.

Elasticsearch

Who: someone with the manage_security privilege on the cluster (on Elastic Cloud, a deployment admin).

  1. Copy the Elasticsearch endpoint. In the Elastic Cloud console, open the deployment and click Copy endpoint next to Elasticsearch. It has .es. in it. The Kibana address (with .kb.) does not work.

  2. Create an API key that may only read the index. In Kibana, Stack Management → API keys → Create API key, name it "SourceLace", turn on Control security privileges, and give it read on the one index, for example:

    {"sourcelace": {"indices": [{"names": ["docs"], "privileges": ["read", "view_index_metadata"]}]}}
    

    view_index_metadata lets Test connection check the field names. Copy the key's Encoded value: it is shown once, when the key is created. The key's id and its api_key part on their own do not work.

  3. Note the field names. In Kibana, Stack Management → Index Management, open the index and look at Mappings. SourceLace's defaults are title, url, content and modified. Indexes filled by FSCrawler use file.filename, path.virtual (or file.url), content and file.last_modified; Elastic's web crawler uses title, url, body_content and last_crawled_at.

  4. For a shared login instead: create a user whose role has read (and view_index_metadata) on the index.

  5. For each person's own account: your cluster must accept tokens from your identity provider (a JWT or OpenID Connect realm). Register SourceLace there as an OAuth client with redirect URL https://mcp.sourcelace.ai/connect/callback, and give the SourceLace admin its authorization and token addresses, client ID and scopes.

Elastic's own guide: Elastic documentation.

OpenSearch

Who: a security admin of the cluster (the all_access or security_manager role).

  1. In OpenSearch Dashboards, Security → Roles → Create role: index permissions read and indices:admin/mappings/get on the one index.
  2. Security → Internal users (or your identity provider's backend roles): create a user for SourceLace and map it to that role, or map the people's own roles for each person's sign-in through OpenID Connect or JWT.
  3. Note the field names from the index's mapping, as for Elasticsearch.

OpenSearch's own guide: OpenSearch documentation.

Network, every service: if the service accepts only some IP addresses, allow SourceLace's (ask support for the list).

Check that it worked: the SourceLace admin saves the source and presses Test connection. With a shared account, it searches once and checks that the index has the fields the source names. With each person's own account, one person clicks Connect and asks for a file they know is in the index.

2. Add the source (SourceLace admin)

Option Type Default Example What it is
service Choice (required) azure_ai_search elasticsearch, opensearch or azure_ai_search.
endpoint Text (required) https://corvanta.search.windows.net The service's address, starting with https://.
index Text (required) contracts The one index this source searches.
auth Choice (required) oauth How people sign in: oauth, api_key or basic (see above).
secret Secret (none) The API key, query key or password, for a shared account. Stored encrypted and never shown again.
username Text (none) svc-sourcelace The user name, with basic.
client_id Text (none) 1a2b3c4d-... With oauth: the OAuth client registered for SourceLace.
client_secret Secret (none) Its client secret, for a confidential client. Stored encrypted and never shown again.
tenant Text (required with oauth) corvanta.onmicrosoft.com Azure AI Search with oauth: your Microsoft Entra tenant's ID or domain (Entra admin center, Overview, Tenant ID).
authorize_url, token_url, scopes Text Filled in for Azure AI Search Elasticsearch and OpenSearch with oauth: your identity provider's addresses and scopes.
title_field Text title (Azure AI Search: metadata_storage_name) file.filename The field that holds each file's title or name.
link_field Text url (Azure AI Search: metadata_storage_path) path.real The field that holds where the file is kept. Shown as a link when it is a web address.
content_field Text content content The field that holds the text to search and show passages from.
modified_field Text modified (Azure AI Search: metadata_storage_last_modified) file.last_modified The field that holds when the file last changed, for searches by date.

Then press Test connection: with a shared account, SourceLace searches the index once, says whether it answered, and checks that the index has the title, link, text and date fields the source names.

How people use it

People ask in plain words ("Find the Corvanta renewal contract in the contracts index"). Search index sources take the files query language with one action, search:

{"action": "search", "text": "corvanta renewal", "modified_after": "2026-09-01", "limit": 20}

Each hit has the file's name, a link back to it, its location as the index gives it, when it last modified, a snippet showing where the words are, and the index's own score. Add "match": "name" to search titles only. Search index sources are also searched by Searching everything, where it is on. Each search is in your audit trail, with how many hits it found; never their text.

When something goes wrong

What you see What to do
"Set the option service: elasticsearch, opensearch or azure_ai_search ..." (or endpoint, index, auth) Fill in that option on Connections.
"OpenSearch has no API keys: choose basic (a shared login) or oauth." Choose another way to sign in.
"Azure AI Search takes no login: choose api_key (a query key) or oauth (Microsoft Entra ID)." Choose another way to sign in.
"... refused the search: your account may not search the index ..." Ask your search service's admin for read access to the index (Azure AI Search: Search Index Data Reader).
"... refused the search: the shared account the admin set may not search the index ..." The key or login is wrong or cannot read the index. Enter a new one on Connections.
"... has no index called ...." Check the index option.
"... could not search (...)" The service's own reason follows, such as a field that does not exist. Check the field options.
"That is Kibana's address. Use the Elasticsearch endpoint instead ..." or "... answered with a redirect, which SourceLace never follows. ..." Enter the Elasticsearch endpoint: Elastic Cloud console, the deployment, Copy endpoint next to Elasticsearch.
"Set Microsoft Entra tenant: your tenant's id or domain ..." Enter the Directory (tenant) ID from the app's Overview in the Entra admin center.
"Azure AI Search refused your sign-in. In the Azure portal, ..." Do steps 1 and 2 of Azure AI Search, then connect again.
"... did not accept the API key. Paste the key's Encoded value ..." Create a new key and paste its Encoded value (step 2 of Elasticsearch).
"... lacks the read privilege on the index .... Give its role read on ...." Give the key's, user's or person's role read on the index.
"The index has no field ... (the text to search), so searches find nothing. Text fields it has: ..." Set Text field to one of the fields listed.
"The index has no field ..., so Title field shows nothing. ..." (or Link field, Date field) Set that option to one of the fields listed.
"The index's list of fields could not be read with this key, so the field options were not checked." Nothing is wrong. To have them checked, give the key view_index_metadata (Elasticsearch); an Azure query key can never read them.
"... is busy. Try again in a moment." Your search service asked SourceLace to slow down. Try again shortly.
"A search index source only searches ..." Ask for a search. To read a file, open its link, or read it from the file source that holds it.
"... sent back something other than search results." The address answered with something that is not a search service (such as a web page).