If you have connected apps (see Connected apps overview), you can let Ask use them in a conversation. Signed-in Plus, Pro and Business users can switch Tools on per conversation using the bar above the composer, which lists the active apps. Free and anonymous users get no tools.
How it works
Tool use is a short pre-stage before the normal engine runs. A small planner model sees only your question, the recent turns and a compact list of allowed tools (numbered, never URLs or secrets). It may propose calls as structured data. Keplar's server runs them, and the results are handed to the answering models as context. Simple questions never trigger tools: a deterministic gate must pass before the planner is even called.
What you see
States like "Using <app>: <tool>" with receipts (status, time, call id), approval cards, and refusals. The planner's private reasoning is never shown or stored; only the proposal that was run or put on hold.
Approvals
Reading tools can run automatically only if all four hold: the tool is marked read-only, it is set to auto-run, it has been granted to Ask, and your account setting "Let Ask run read-only tools without asking" is on (it is off by default). Everything else, including every write, every unknown tool and every scheduled run, produces an approval card in the chat. Nothing runs until you press Approve.
Limits
At most 5 tool steps per turn, 3 planner rounds, 3 calls per round and 45 seconds in total. Identical calls are de-duplicated, and tool results are capped in size. Hitting a cap ends tool use and the answer continues without further tools.
Cost
The planner call is metered, and each tool call costs about one credit. Tool results reduce the remaining answer budget.
Treat results as data
Tool output is sanitized, fenced and labeled as data, not instructions, before the models see it. It is sent to the models answering that question, which the interface discloses. Do not connect an app whose data you would not want a model provider to receive.
Related
- Connected apps (MCP) overview: Connect remote Model Context Protocol servers so Keplar and its agents can use outside tools, with per-tool permissions, approvals and plan limits.
- Tool permissions and approvals: Read versus write tools, grants, auto-run rules, approval cards, schema-change locks and how scheduled runs always ask first.
- Add a connection: Connect an MCP server step by step, test it with a handshake, enable only the tools you want, and the address and credential rules that apply.