Senso Documentation — The Trusted AI Answer

One person on your team uses Claude Code. Another uses Codex. Another uses Cursor.

Each is researching, planning and building with an agent, but much of what they learn stays in separate conversations. A decision gets explained again. A correction never reaches the next person. An unanswered question gets rediscovered.

Shared context is knowledge that people and their agents retrieve, use and improve together, across tasks and conversations. In Senso it is your knowledge base: everything your team saves, every agent can find, whichever tool it runs in.

How work builds on work

Take product planning:

A product lead saves the requirements and the reasoning behind them.
A developer's agent searches Senso before starting, and builds from that plan.
It hits a question the plan does not answer: what happens to existing customers? The team decides, and saves the decision.
Another teammate's agent finds that decision when it writes the tests and the release notes.

That is the cycle: save what you learn, start from what the team knows, settle what is missing, and the next task starts further ahead. It works the same for coding, research, support answers or anything else a team does with agents.

Run it with your agent

The agent skills run this for you. Install them, then ask your agent for Shared Context:

bash

senso setup

Let's go through Shared Context.

The first time, if your knowledge base is empty, your agent helps you add one document: a short note on what your team does and what its agents should know, or a file you already have.

Save what you learn

Save the things a teammate's agent would otherwise have to rediscover: requirements, decisions, constraints, findings, and the reasoning behind them.

Save a decision with what the next person needs. The decision, why it was made, what is still open, and the next step:

For example, a note titled Existing-customer pricing on the new tiers:

Decision: existing customers keep their current price for the rest of their term, then move to list price at renewal.
Why: the cost of keeping their price is small, and the risk of losing them mid-term is not.
Open: whether annual customers get 30 days' notice before renewal. Legal to confirm.
Next: Dana updates the pricing page by Friday.
Owner: Dana · Approved 2026-10-01

Write who owns it, and whether it is settled, into the note itself, as in the last line above. Senso tells every agent where an answer came from. Whether to rely on it is the owner's call, and that line records the call where every reader will see it: it comes back with the note in any search, whichever tool the agent runs in.

The Senso skills also tag what the team saves, so you can list and filter by status and owner:

Tag	Means
`status:approved`	Settled by the person who owns it.
`status:draft`	Not settled yet. Treat it as a lead.
`owner:<name>`	Who to ask, and who can change it.
`decided:<date>`	When it was decided.

A document without these tags, like a page imported from your website, is neither approved nor anyone's. The skills keep what the team settles in a shared-context folder unless you already keep it somewhere else. Senso adds its own topic tags as well, and leaves yours alone.

Start from what the team knows

Every agent on your team searches the same knowledge base, so a teammate's saved work is there the moment it is processed. Nothing pushes it into a conversation that is already running: ask your agent to search Senso before it starts a task, or start a fresh session.

bash

senso search "What did we decide about pricing for existing customers?"

Each answer cites the documents it came from, so you can see whose work you are building on, and the owner and status line in each note tells you whether it is settled.

To point a teammate at something specific, send them the document's name and its kb_node_id, and say what you want them to do with it: read it, use it for a task, correct it, or add what is missing. They open it, tags included, with:

bash

senso kb get <kb_node_id>
senso kb get-content <kb_node_id>

Settle what nobody has answered

When an agent searches Senso and finds nothing, Senso records the question. The list of these open questions is your team's to-do list for its shared knowledge:

bash

senso gaps list

A question asked once from an agent or the CLI is held back as weak, and appears in the list when it is asked again, so a single test search does not crowd out real demand. Keep your own tests out entirely with --no-gap-signals.
The list also includes claims an eval could not back up. The commands below show how to list only the questions people asked.
Only a search that finds nothing at all is recorded. If a related document comes back, the question is not listed, even when the answer is wrong. Asked "Do monthly customers get a loyalty discount at renewal?" with only a pricing decision on file, an agent was told "No": a guess from what the note did not say. Ask your agent to check what it is told about the things that matter, and correct what is wrong.

To settle one:

Check what agents are told today. Search the question, and read the answer and its sources.
Ask who owns it. If it is not you, leave it open for them.
Write it down with its conditions. "Existing customers keep their price" is not enough; until when, and with what exceptions? A fact saved without its conditions is the next wrong answer.
Mark it approved, or save it as a draft if it is not settled yet, with the status written into the note.
Search again. The answer should now cite the new note, and the question closes on its own.

Revise as things change

Editing a document saves a new version, and earlier versions can still be read:

bash

senso kb get-content <kb_node_id> --rev 1

Keep it dependable

Sharing more only helps if people can tell what to trust.

Save what a person said and approved, not what a model generated. Feeding every generated answer back into the knowledge base piles up errors.
Investigate before declaring something missing. A search can miss a document that exists. Check whether it is there under another name before writing a second one.
Tell a missing fact from a wrong one. Nothing on file is a gap. When two documents disagree, that is a conflict, and someone has to decide which is right.
Check generated work against its sources. An eval checks each claim in a draft against your knowledge base before anyone relies on it.
Give people and agents access to what they need. Permissions decide who can see which folders.

Commands

The commands behind the cycle, for running it yourself:

bash

# Save a note, then tag it
senso kb create-raw --data '{"title": "...", "text": "...", "kb_folder_node_id": "<folderId>"}'
senso kb tags set <kb_node_id> --names "status:approved,owner:Dana,decided:2026-10-01"

# Start from what the team knows
senso search "<your question>"

# The open questions
senso gaps list

# Only questions people asked, including ones asked once, most-asked first
senso gaps list --origin api_unanswered_question,unanswered_question,documents_didnt_answer,flagged_answer --status weak,open,reopened --sort demand

# Revise a document, and read an earlier version
senso kb patch-raw <kb_node_id> --data '{"text": "..."}'
senso kb get-content <kb_node_id> --rev 1

Every command is in the CLI reference.

Next steps

Verification Loop — find the AI questions you are missing from, and publish a checked answer.
Connect Your Agent — set up Senso for your agent.
Permissions — who can see which folders.