Your product knowledge, on tap
Build Karumi agents and read their sessions from any AI tool.
Karumi runs as an MCP server.
What it is
One connection, two jobs
Karumi exposes your product knowledge and your agents over MCP. Connect it once, and any AI client can build agents against your product or read back what those agents learned.
Setup
Point your client at it
Set up agents for a specific job:
One agent for each job:
-
Sales
One per prospect, briefed on the account.
-
Marketing
One per campaign, matched to the message.
-
Support
One per customer, scoped to what they use.
Agents become disposable. One per campaign, one per prospect — generated rather than configured.
Read back what they learned
-
Identify patterns
What's being asked, repeatedly, across every agent you run.
-
Evaluate agents
Score them. Find the ones answering badly.
Launch follow-ups. Act on what a session surfaced, without a human reading it first.
Tools
The tools available
| Tool | Description |
|---|---|
| Create agents | |
create_agent | Create an agent scoped to a job type, briefed on a given account or context. |
list_agents | List every agent you've created and its current status. |
update_agent | Change an agent's scope, briefing, or the context it's tied to. |
delete_agent | Remove an agent once its job is done. |
| Extract insights | |
list_sessions | List sessions across every agent, filterable by agent, date, or signal. |
get_session | Read a session in full — the transcript, and what it surfaced. |
search_sessions | Search across sessions for a pattern or a repeated question. |
evaluate_agent | Score an agent's answers and find the ones answering badly. |
create_followup | Launch a follow-up action from what a session surfaced. |
Use cases
What this is for
- 01
An agent per prospect, generated from the CRM the night before the call.
- 02
Weekly pattern extraction across every session, straight into the roadmap.
- 03
Automatic follow-up the moment a session surfaces a buying signal.
