Common workflows
Eight recipes that go end to end. Each one names the surfaces involved so you can see how the pieces fit together.
Trace one governed path from your request to a source, cited answer, artifact, or durable Task result.
1. Answer a question you cannot find the answer to
Search → preview → Chat
Search first. If the top cards look right but you need them synthesized, switch to Chat and ask the same question — the assistant retrieves the same authorized sources and writes a cited answer. Click a citation to land on the exact passage.
Use Search when you want the original. Use Chat when you want the conclusion.
2. Ground an answer in a file you have
Documents → upload → Chat
Upload the file in Documents. Kaizen stores the original, extracts its text (including OCR for scanned pages), and indexes it privately to you. Then ask about it in Chat — small files can be read whole, large ones are retrieved in the relevant parts. Either way the answer cites the passages it used.
Your uploads are never mixed into the shared corpus.
3. Keep a topic's sources together
Collections → scope a chat
Create a collection and add the sources that matter for a recurring topic. Then point a chat — or a custom agent — at that collection. Retrieval stays inside it, which makes answers tighter and more predictable.
Sharing a collection never widens access: each recipient's own permissions are re-checked against its contents.
4. Stop retyping the same request
Prompts → save → reuse
When you have refined a request worth keeping, save it as a Prompt. Pull it into any chat and send it as-is or tweak it. When the work is genuinely multi-step and you run it often, build a multi-step Prompt so each step feeds the next.
For one-off questions, do not bother saving anything.
5. Build an assistant for a recurring job
Agent Studio → test → share
Walk the guided builder: name and purpose, persona, tools, knowledge scope, then the live Test step where you have a real conversation with the draft. Adjust until it behaves, then Save & finish — it is live and private to you immediately.
Share it directly with named colleagues when it is ready. Grant only the tools it actually needs; fewer tools means more predictable behavior.
6. Work on something with a colleague
Chat → Share → one room
Open a saved chat, choose Share chat, and enter the colleague's verified email. Review the access summary — it tells you exactly what travels: the retained history, linked documents, artifacts, citations, and tool results.
What does not travel: your sandbox, files, installed packages, environment values, and credentials. Each turn runs under whoever sent it.
7. Organize a body of work
Projects → add chats and Library items
Create a Project, then add existing chats and link Agents, Artifacts, Prompts, Documents, Collections, or Tasks. A Project points at existing items — it never copies anything, and unlinking or deleting the Project destroys nothing.
Invite named Project members when the whole body of work is shared, rather than sharing each chat one at a time.
8. Make something happen on a schedule
Prompts → Tasks → activate
Tasks are off in this release, so by default Tasks is not in your rail and this workflow is unavailable. If your administrator has turned Tasks on and the entry is still missing, Kamiwaza is not reporting the readiness Tasks need.
Save the work as a Prompt first. Then create a Task pointing at that exact version, choose a schedule, and review the server-authoritative preview: next three occurrences, destination, effective limits, tools, and approval behavior.
Run the draft. Only a successful test against the current revision unlocks Turn on task. After that it runs without an open browser, and the run trace tells you what actually happened.
You can also just ask a chat agent: "create a draft called Monday update using my Weekly summary prompt." Same service, same owner binding, same audit trail.