How to ask another mm agent from your session

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61

# How to ask another mm agent from your session

This guide shows how to put a question to another mm agent — another model, another system prompt, other tools — and keep its answer in your conversation. It assumes you already run `mm` with a working provider.

## Steps

1. Write (or reuse) one configuration file per agent you want to ask. Each is an ordinary mm configuration; `demos/llmman/` ships two, for llmman:

   - `demos/llmman/agent.jan-nano.yaml` — `hf.co/Menlo/Jan-nano-gguf:Q4_K_M`, named Jan;
   - `demos/llmman/agent.lucy.yaml` — `hf.co/Menlo/Lucy-gguf:Q4_K_M`, named Lucy.

2. Declare them in YOUR configuration, under `mmConfigs`. Paths are relative to the directory of the file you write them in:

   ```yaml
   mmConfigs:
     jan: agent.jan-nano.yaml
     lucy: agent.lucy.yaml
   ```

   `demos/llmman/agent.llmman.yaml` already does this.

3. Start `mm` with that configuration:

   ```bash
   cd demos/llmman
   ../../mm agent.llmman.yaml
   ```

   An agent whose file is missing shows up as a `[warning: mmConfigs: …]` line; the others work.

4. Ask a question:

   ```text
   > /ask-jan In one short sentence: what is the capital of France?
   🗣 asking jan (llmman/hf.co/Menlo/Jan-nano-gguf:Q4_K_M)…
   The capital of France is Paris.
   🗣 jan's answer is in the history for your next question
   ```

   The first ask of an agent starts it, so it takes a little longer; the next ones reuse it.

5. Use the answer: your next question goes to your own model, which now sees Jan's answer.

   ```text
   > What did Jan answer?
   🤖 Jan answered that the capital of France is Paris.
   ```

## Variants

- **Let the other agent see the conversation.** Use `/ask-with-context-lucy <prompt>`: Lucy receives a transcript of the conversation (questions, answers, one line per tool call — never tool outputs) before your question.
- **In the terminal or the window, by name.** `/ask lucy <prompt>` and `/ask-with-context lucy <prompt>` are the same commands with the agent as an argument.
- **In an editor (ACP).** The menu lists `/ask-jan`, `/ask-with-context-jan`, `/ask-lucy`, `/ask-with-context-lucy`; when Lucy wants to run a command, your agent's `yolo` decides: with `yolo: true` it runs, with `yolo: false` the editor's own permission dialog asks you — whatever Lucy's own file says. A `yolo` in Lucy's entry of YOUR `mmConfigs` (`lucy: {path: agent.lucy.yaml, yolo: true}`) decides for her instead. See [How to confirm tool calls](confirm-tool-calls.md).
- **Stop a long answer.** `/abort`, `Esc` or `Ctrl+C` — nothing is kept from a cancelled ask.
- **Agents on other providers.** Any mm configuration works: an Ollama agent next to an llmman one is just another `mmConfigs` entry.

## See also

- Every form, message and error: [slash commands reference](../reference/slash-commands.md#asking-another-agent)
- The `mmConfigs` key: [configuration reference](../reference/configuration.md)
- Why another agent is a separate process spoken to over ACP: [asking another agent — explanation](../explanation/asking-another-agent.md)