Teams whose know-how is spread across PDFs, an intranet and people's heads.
What it costs today
People can't find the policy, so they ask a colleague, who stops what they're doing to search for it, or they guess.
What solved looks like
Anyone asks in their own words and gets the answer with the document and page it came from, and "I don't know" when it isn't written down.
Answers grounded in your documents, each with its source.
"I don't know" instead of a guess when nothing relevant is found.
One place to fix: update a document and the answers change.
How it runs
Internal knowledge assistant step by step
Each stage is a real step in SentientOne. Timings are medians from our latest benchmark; your model and network change them.
Internal knowledge flow. A colleague asks in your intranet tool, which calls the REST API. 1: the key and the agent are checked and one credit is reserved. 2: a repeated first question comes from the answer cache. 3: your policy PDFs, handbook pages and snippets are searched by meaning and keyword and reranked. 4: only passages scoring at least 0.30 go to the model, with a rule to answer only from them. 5: the model writes the answer, or says it doesn't know. 6: the answer streams back with its sources.
Channel
Your intranet tool or an internal app, through the REST API
Question
“How many days of parental leave do I get?”
Context
conversation_id for follow-ups
Agent: Organisation Knowledge
X-Request-Id
SENTIENTONE
Auth, credits
Preflight
Your secret key from your intranet's server; 1 credit reserved.
60 req/min · 402 at zero
Answer cache
~46 ms
The same first question to an unchanged agent comes back instantly.
15-minute TTL · cleared on upload
Knowledge
~450 ms
Policy PDFs, handbook pages and snippets, by meaning and exact wording.
leave-policy.pdf · handbook · snippets
Grounding
No guessing
Only passages scoring 0.30 or more go in, with a rule to answer only from them.
min score 0.30 · top 5
Generation
Cited
Answers from the passages, or says it isn't written down.
GPT-5.5 on your key
Streaming
OTel trace
The sources event first, then the answer as it's written.
text/event-stream
Answer
Done · 1 credit
You get 18 weeks of paid parental leave, which you can start up to 4 weeks before the due date [1]. Tell your manager at least 8 weeks ahead [2].
1Parental leave · leave-policy.pdf, p. 3
2Requesting leave · handbook/leave
trace · 2 sources · 1 credit
Channel
Your intranet tool or an internal app, through the REST API
Question
“How many days of parental leave do I get?”
Context
conversation_id for follow-ups
Agent: Organisation Knowledge
X-Request-Id
question
SENTIENTONE
Auth, credits
Preflight
Your secret key from your intranet's server; 1 credit reserved.
60 req/min · 402 at zero
credit
Answer cache
~46 ms
The same first question to an unchanged agent comes back instantly.
15-minute TTL · cleared on upload
miss
Knowledge
~450 ms
Policy PDFs, handbook pages and snippets, by meaning and exact wording.
leave-policy.pdf · handbook · snippets
top 5 chunks
Grounding
No guessing
Only passages scoring 0.30 or more go in, with a rule to answer only from them.
min score 0.30 · top 5
grounded
Generation
Cited
Answers from the passages, or says it isn't written down.
GPT-5.5 on your key
tokens
Streaming
OTel trace
The sources event first, then the answer as it's written.
text/event-stream
cache hit · first token ~46 ms
first token ~1.1 s with a knowledge search
answer + sources
Answer
Done · 1 credit
You get 18 weeks of paid parental leave, which you can start up to 4 weeks before the due date [1]. Tell your manager at least 8 weeks ahead [2].
1Parental leave · leave-policy.pdf, p. 3
2Requesting leave · handbook/leave
trace · 2 sources · 1 credit
Set it up
The walkthrough, in 7 steps
1
Add your LLM key
Agents run on your own model provider key; there's no shared one, so nothing answers until you add it. Create a key with OpenAI, Anthropic, Google Gemini or Groq, then open Configure → LLM keys → Add key. It's encrypted at rest with AES-256-GCM, used only to call your provider, and never shown again in full. If a reply fails with INTERNAL_ERROR straight away, the agent has no key; with LLM_AUTH_ERROR, the provider rejected it.
Tell it who it serves, to cite the document, and to say clearly when something isn't covered.
System prompt
You answer staff questions about our policies and processes.Answer only from the knowledge base, and name the document you used.If the answer isn't in the knowledge base, say so and suggest asking People & Culture.Don't give legal or medical advice. Keep answers short.
Knowledge → Documents: upload a dozen key policy PDFs, about 60 pages in all (300 credits; up to 50 MB each; scanned image-only PDFs have no text to index). Add 20 one-line Snippets for the facts people ask most (20 credits).
Call the API from your intranet's server (never the browser) and show the sources under each answer.
Your intranet's server
curl-N https://api.sentientone.ai/v1/chat/stream \-H"Content-Type: application/json" \-H"X-Api-Key: $SENTIENTONE_API_KEY" \-H"X-Agent-Id: $AGENT_ID" \-H"X-Request-Id: $YOUR_REQUEST_ID" \-d'{"message": "How many days of parental leave do I get?"}'
Open Tracing to see every request: what knowledge it found, each tool call, tokens and timing, and the error code if it failed. Analytics charts requests, errors and response time per agent. Under Billing, turn on auto top-up so the agent never stops for want of credits.