Niches · Company knowledge & data catalogues
Answer a question from company documents and cite the source
No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.
What counts as done
An answer whose every claim links to a source document (and page or message where the source is long) the user can open.
Measured, for our own tests: the answer comes in one call within 15 seconds, and a question with no supporting document returns no answer rather than an uncited one. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: what does our travel policy say about X
, where are the wire transfer instructions
, what did we decide about this in Slack
, find the doc that explains this
Does the job, with a limit 10
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Atlan C Atlan Pte. Ltd. | search_knowledge returns passages citing file name, folder, page and title, and ask gives a grounded answer, but only over files uploaded to Atlan, the knowledge tools are early preview and may be off, and ask reasons before answering with no time stated. source |
| Dust B Permutation Labs | Data source search returns documents with source_url, chunks and timestamp, and the MCP server searches Spaces as the user. But workspace search streams nodes, not cited answers, unthreaded Slack links point to a chunk window rather than the message, and no latency is stated. source |
| Gemini Enterprise B Google LLC | StreamAssist answers carry grounding references with uri, title and pageIdentifier, and segments mapping text to references. No latency figure or no-answer behaviour is documented, assistant use is capped per licence per day, and acc-36 could not get users connected to Confluence. source |
| Glean B Glean Technologies, Inc. | Chat returns ChatMessageCitation with the source document and referenceRanges of snippets per text range, and search and chat are MCP tools. No latency figure, chat is capped near 0.5 calls a second, no documented no-answer behaviour, and access starts with a sales call. source |
| Guru D Guru Technologies, Inc. | Ask returns an answer with a sources array of cards and source documents (title, URL, last modified, verification state), a status that can be ERRORED, and a per-agent noAnswerMessage. Sources are per answer with no page or passage position, and no latency is stated. source |
| Hexis Cloud D Bevelites GmbH | Agents grep (ripgrep) the git knowledge base and read_file pages, then answer themselves, so citations are whatever file paths the agent gives. It covers only pages in the Hexis repository, takes several calls, and has no no-answer signal. source |
| Onyx B DanswerAI, Inc. | search_indexed_documents returns title, url, source_type, content and updated_at, and a support team used it daily (acc-07). Links are file-level only (acc-21), chat has returned empty answers with correct sources (acc-25), and MCP returns results, not a cited answer. source |
| OpenMetadata C Collate, Inc. | company_context returns question-and-answer pills with their source file, and find_context searches Context Center articles, but only for files loaded into Context Center, with no page position, and the query path needs vector embeddings (it returned nothing on Elasticsearch, issue 36463). source |
| Overclock F Overclock Technologies Limited | find_evidence returns context with evidence items (quote, document title, page when known, document id) and answerable false when documents don't cover the question. Results carry ids, not a link; the uri needs get_document, and no latency is stated beyond 'fast'. source |
| Lens (Thought Industries) F Thought Industries, Inc. | POST /v1/ask returns an answer with a sources array (title, excerpt, URL when available), and the FAQ says it declines rather than guesses. Sources are per answer, not per claim, carry no page position, and no latency is stated. source |
Couldn't check 2
The docs were closed to us, or silent. These carry no mark either way.
| Listing | Why |
|---|---|
| Cohere North D Cohere Inc. | private.docs.cohere.com disallows all crawling in robots.txt, so the chat and search response schemas went unread; the product page claims search and retrieval but shows no citation fields. |
| Secoda E Secoda, Inc. | search_documentation searches Secoda documents, questions and glossary entries, but the local server only proxies to Secoda's API, so the result fields (source link, position) are not in the repo; Secoda's docs were not re-read because its Terms of Use forbid bots. source |
Doesn't do this job 8
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| Alation C Alation, Inc. | A data catalogue; the SDK and MCP tools answer catalogue, lineage, data quality and data product questions, not questions over wiki pages, drive files or chat. source |
| Basedash MCP B BaseDash Inc. (Delaware, United States) | A BI analyst over databases, warehouses and SaaS data, not a search over wiki pages, drives or chat; ask_question answers from SQL results, not from documents. source |
| ClicData MCP F ClicData SAS | ClicData MCP answers data questions from ClicData's warehouse with charts and dashboards; it does not search wiki pages, drives or chat for cited answers. source |
| Collibra B Collibra Inc. | A data governance catalogue, not a document search: no connector to wikis, drives or chat and no cited-answer tool. discover_business_glossary answers about glossary terms only. source |
| DataHub C Acryl Data, Inc. (d/b/a DataHub) | A data catalogue, not a search over wiki, drive and chat. search_documents and grep_documents cover only documents saved into DataHub, and the server returns entities with URLs rather than a cited answer. source |
| Marmot B Marmot Data Ltd | A data catalogue; its MCP tools search assets, glossary terms, teams and data products, not wiki pages, drive files or chat threads. source |
| Precisely AI Studio F Precisely Incorporated | AI Studio is a catalogue of ready-made apps, agents and skills for data integration, validation, enrichment and location tasks, run on sample data; no search or cited answers over company documents is described. source |
| Procore AI Go F Procore Technologies, Inc. | The app answers a person's spoken questions from Procore project records with links to the items, but there is no API, MCP server or other interface an agent can call. source |
What stops agents finishing
- no self-serve route: Glean, Guru, Cohere North, Overclock and the hosted catalogues start with a sales call
- source app's API terms forbid indexing (Slack, May 2025)
- answer synthesis returns empty with the right sources (Onyx 14333)
Around it
Before: chat.user-question, code.agent-task. After: support.reply-ticket, work-docs.draft.
Where this comes from
Weight 26 of 100 in Company knowledge & data catalogues: 12 of 38 counted accounts; the job most often described (Onyx support-team use, Glean search API and MCP from Cursor, Grapevine and context-layer threads)
The same as data: /api/v1/use-cases/company-knowledge.cited-answer.json, or the MCP tool get_use_case with id company-knowledge.cited-answer. Other jobs in this category: /api/v1/niches?category=company-knowledge.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.