Subagents

My Agents turns other agents into context for DeepTutor. It does two distinct things: it lets you consult a live agent (a Claude Code, Codex, Gemini, Kimi, opencode, or MiMo Code CLI running on your machine, or one of your Partners) from inside a chat turn, and it lets you import their past conversations as named, searchable, resumable agents.

Where it is

Open My Agents in the left sidebar. From a chat turn you can also reach the live agents two ways: the Agent chip in the composer toolbar (the robot icon), or by typing @ in the message box.

The My Agents page: Connected agents on top, Imported conversations below

The page has two sections that mirror the two ideas.

Connected agents — consult a live agent

A connected agent is a real agent DeepTutor can talk to in real time. Press Connect agent and pick one of:

a coding CLI on this machine — Claude Code, Codex, Gemini, Kimi, opencode, or MiMo Code, or
one of your Partners.

When you consult a connected agent, DeepTutor does not paste a transcript — it actually runs the other agent and streams its work back. Under the hood a connected agent is a third kind of context source with one exclusive power: the chat loop’s consult_subagent tool drives the agent over multiple rounds and shows its full run live in the Activity panel.

Consulting from chat

Select the agent with the Agent chip, then set Max rounds DeepTutor may ask — the ceiling on how many back-and-forth rounds the consult may take.

The Agent chip: pick a connected agent and set max consult rounds

Or just type @ to mention an agent inline for a single turn.

Typing @ opens "Talk to an agent" inline

What a live consult looks like

Consult a Claude Code agent and it runs against its own working directory — reading files, grepping, and reasoning over the real repository. Its tool calls stream into the Activity panel while DeepTutor folds the conclusion back into your answer.

Consulting a Claude Code subagent: its file reads and greps stream live into the Activity panel

Consult a Partner and DeepTutor sends the question into that Partner’s own session — so the Partner answers with its soul, its library, its private memory tools (partner_search, partner_read), and its skills (read_skill). One chat thread stays bound to one Partner session, so follow-ups continue the same conversation.

Consulting the "Panda Kate" Partner: it answers using its own memory and skills

Imported conversations — agents from your history

The second half of the page imports your existing Claude Code and Codex conversations and presents each source as a named agent you can search, open, keep chatting with, or reference from chat.

Press Add agent to bring in history. Name the agent, then pick which days to import — refreshing later re-syncs those days and pulls in any new conversations on them. This is selective on purpose: you choose exactly which slices of your terminal history become DeepTutor context.

The Add agent dialog: name the agent and select days to import

Imported agents show their source, conversation count, and last sync time. From there you can:

Open a conversation to keep chatting with it inside DeepTutor.
Refresh an agent to pull in conversations from new days.
Reference an imported conversation in any chat turn via + → My Agents — DeepTutor reads it as a third-party transcript (it stays “their” conversation, not DeepTutor’s own voice).

How the two halves differ

	Connected agents	Imported conversations
What it is	A live agent DeepTutor runs and talks to	A read-only history brought in as a named agent
Source	Local coding CLI (Claude Code / Codex / Gemini / Kimi / opencode / MiMo), or a Partner	Past Claude Code / Codex sessions
In chat	Consult via the Agent chip or `@`; runs streamed live	Reference via `+` → My Agents, or open to continue
Freshness	Real-time, every consult	Snapshot; Refresh to re-sync selected days