Senso
Sign inSign up

Step 1 of 4 · Stage 01

Build the Context Layer

Register what you own, then get your sources into a knowledge base your agents can query.

01Build the Context LayerGround agents in your own sources02Evaluate & RemediateCompare model answers with your sources03Generate & VerifyDraft from your sources, then verify it04Publish & ObservePublish it, then measure what changed

A knowledge base your agents query: send a question, get an answer plus the ids of the documents behind it. You write to it here, and step 2 corrects it, so its sources stay right as the loop runs.

  1. Register your websiteit decides which citations count as yours
  2. Populate the context layerat least one document has to exist
  3. Choose the right sectionit decides what the source is checked against
  4. Confirm it answersprocessing has to finish first

Produces a queryable knowledge base, written back into the context layer.

How to run it

How to run this step, in order: the step, the API call it maps to, and why
StepAPI callWhy it matters
Register your websitePUT/org/meReplaces the organization’s website list. A cited URL is classified as your own source when its domain is on that list, so until one is registered nothing can come back as yours however well the content performs. The Brand Kit carries a domain too, but that one is descriptive — this is the list citations are classified against.
Populate the context layerPOST/org/kb/uploadUploads files into the knowledge base. Nothing downstream can run until something is queryable, so get one real folder in before you plan the structure — structure is easier to fix later than absence is. Markdown and plain text go in through POST /org/kb/raw instead, or over MCP; the upload call rejects .md files.
Choose the right sectionPOST/org/kb/foldersCreates the folder a source belongs in. Where a source lives decides what it can be checked against later, which makes the section a decision about evidence rather than about tidiness.
Confirm it answersPOST/org/searchQuery it and read what comes back: an answer plus the ids of the documents behind it. A document is not searchable the moment it lands, so poll GET /org/kb/nodes/{id} until content.processing_status reads complete — roughly every 15 seconds — rather than assuming the upload was the last step. GET /org/kb/sync-status answers a different question: whether queued moves and deletes are still propagating org-wide.

Ask your agent

Ask for our website and register it, then bring our key documents into the knowledge base and put each one in the section it will be checked against.

Context Layer

The five sections together. Everything Senso treats as your approved truth, as opposed to anything it reads from the open web.

Full definition →

Set your Brand Kit

Two fields are required: your organization name, and a domain you own. Everything after this is scoped to them. Voice, tone and writing rules go in through the same call and shape what gets drafted in step 3. The domain that decides which citations count as yours is the organization\u2019s registered website, set separately with PUT /org/me.

These two scope everything after them. Classifying a citation as your own source is a separate setting — the organization's registered website list.

bash
curl https://apiv2.senso.ai/api/v1/org/brand-kit \
  -H "X-API-Key: $SENSO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"guidelines": {"brand_name": "Acme Credit Union", "brand_domain": "https://acme.com"}}'

Stays in your browser. Runs once you have a key.

Get one document in

One real document first, before you plan anything else. Nothing downstream can run until something is queryable, and structure is easier to fix later than absence is. Markdown and plain text need the raw call rather than the upload one, which rejects .md files.

Put your sources in the right section

Five sections, not one folder. Where a source goes decides what it can be checked against later. Run a test search to see which sections actually carry an answer.

Brand Kit

/org/brand-kit6 of 6 fields

Voice, tone and rules, so generated content sounds like you.

Knowledge Base

/org/kb14 documents

The sources that make an answer citable.

Content Types

/org/content-types4 templates

Templates controlling structure, tone and CTAs.

Product Catalog

/org/product-lines8 products

What you sell, and on what terms.

Topics

/org/tags11 tags

The tags joining questions, sources and gaps.

Test search

Ask the context layer something and see which sections carry the answer. If one section does all the work, the rest are not pulling their weight yet.

These are documents you own. Whether they are true, current and approved is the gate in stage 03. Counts are illustrative until you connect an organization.

Common mistake

Putting a price in a PDF in the Knowledge Base. A claim about that price cannot then be checked against the Product Catalog, because as far as the system is concerned there is no catalog.

Each has a reference page: Brand Kit, Knowledge Base, Content Types, Product Catalog and Topics.

Ask it something

Query it and read what comes back. You get an answer plus the documents behind it, each with an id you can cite. If nothing relevant is there, it tells you, rather than inventing an answer.

Connect Your Agent connects an agent and runs this whole step in about five minutes.

Grounded answer

An answer built only from your documents, returned with the sources it used. The interesting half is the negative case: no relevant source means no answer, not a guess.

Full definition →

How to reach it

PUT/org/meOpen in API reference →Register the websites you own. This is what classifies a citation as yours.
POST/org/kb/rawOpen in API reference →Ingest markdown or plain text straight into the knowledge base.
POST/org/kb/uploadOpen in API reference →Upload files: PDF, Word, Excel, PowerPoint, HTML, CSV, JSON, XML.
PUT/org/brand-kitOpen in API reference →Set brand name, domain, voice, tone and writing rules.
POST/org/product-linesOpen in API reference →Add a product, its terms, pricing and eligibility. The Product Catalog.
POST/org/tagsOpen in API reference →Create a topic. Tags are how gaps line up with content later.
POST/org/searchOpen in API reference →Query it and get an answer with the documents behind it.

Rules for this step

  • Confirm the organization before any write.
  • Search before ingesting again — identical text does not create a second copy.
  • Updating a document replaces its body. Read it first, send the whole new version.
  • A document is not searchable until compilation finishes. Poll roughly every 15 seconds until it reports complete.
  • Never put API keys, tokens or customer personal data in a source.

No approvals are needed anywhere in this step. Gates start in step 2, when something proposes changing what you consider true; step 3 gates the draft, and step 4 gates anything going out.

Done when

  • Each source is in its own section, not all in one folder
  • A search returns an answer with its sources
  • An irrelevant question returns nothing rather than a guess

An internal answer is not a Verified Source. That takes the claim gate in step 3 and publishing in step 4.

Step 2: Evaluate & RemediateThe Loop
Next step