Your agents stop guessing — they ask, then remember
No knowledge base covers everything. When an agent can't answer from your instructions, business details or knowledge base, it doesn't guess: it tells the contact it's checking, sends you the question, and once you answer it writes back and files the answer away. Ask the same thing tomorrow and the agent already knows.
A closer look at Agent memory
A question it can't answer goes to you, not to the contact as a guess
Turn on "Questions it can't answer" on any agent. When a contact asks something your knowledge doesn't cover, the agent calls its ask-team tool, tells the contact someone will get back to them, and the question lands on your Agent Questions page. Your team gets a bell alert, and an email if you want one.
- No made-up answers to real questions
- The contact is told you're checking
- A red badge and a bell when one is waiting
- Filter by agent and by connected inbox
Answer once, and the agent reuses it on its own
Type the answer and the agent writes back to the contact in its own words — and remembers it. The next time any contact asks something like it, the agent recalls your answer and replies straight away, no re-asking. It works with no knowledge base at all; attach one and the answer goes there too.
- Always-on memory, no knowledge base required
- The agent recalls past answers before asking again
- Optionally saved to the knowledge base as well
- Each answer reused across every conversation
Watch each answer earn its keep
Every answered question shows how many times the agent has reused it and into which conversations, so you can see which answers do the heavy lifting and which gaps keep coming up. The whole thing is driveable over MCP, too.
- "Reused N times" on every answer
- A log of which conversations recalled it
- Reopen or edit an answer any time
- list_questions and answer_question over MCP
The details
Always-on
Answers are remembered and reused even with no knowledge base attached.
Per agent
Each agent recalls its own answers; scoped to its workspace and client.
Notified
A red count in the menu and a bell alert when a question is waiting.
Reuse log
See how often each answer was fetched and which conversations used it.
Who it's for
New service, no docs yet
The agent asks about the thing you haven't written up; your answer becomes the agent's answer.
Edge-case pricing
An unusual quote question gets a human answer once, then the agent handles it every time after.
Agencies
Each client's agents build their own answer memory, kept to that client's workspace.
Agent memory: common questions
Does it need a knowledge base?
No. Answers are remembered and reused on their own. If you attach a knowledge base, the answer is also saved there for semantic search.
Will the agent guess when it isn't sure?
No — that's the point. It asks your team and tells the contact it's checking, rather than making something up.
How does it reuse an answer?
Before asking again, the agent searches past answers for a close match and, if one clearly fits, replies from it. You can see every reuse in the log.
Can I answer from my phone or an integration?
Yes. Answer on the Agent Questions page, or over the MCP server and API.
Works hand in hand with
Knowledge bases
Documents chunked and embedded into pgvector, Supabase, Pinecone or Cloudflare Vectorize.
ExploreApprovals & autonomy
Suggest, draft or auto autonomy, an Approvals queue, business hours and follow-up sequences.
ExploreTest & Debug
Sandbox conversations, a full debugger for every reply, and an AI diagnose-and-fix loop.
ExplorePut it to work on your own inbox
Connect an inbox, describe the job, test it in a sandbox and go live with exactly the autonomy you're comfortable with.