Deep research and long tasks

Some work does not fit in one answer. PP Agent has two modes for it: /goal for a long task that continues until it is done, and /research for a structured investigation with sources.

Long tasks with /goal

Type /goal with the outcome you want. The agent keeps working turn after turn until the goal is met, instead of stopping after one reply.

• A banner lets you pause, resume, or clear the goal.
• A checklist stays pinned above the message box so you can see what is finished and what is next.
• The agent can hand up to five read-only research tasks to sub agents that work at the same time.

Use /goal for work such as “prepare the renewal packet from these contracts” or “audit this spreadsheet and write up every inconsistency”.

Approvals still apply

A goal does not bypass the approval gate. Writes and shell commands still ask according to each tool’s policy, and a scheduled run skips anything that needs approval because nobody is there to approve it.

Deep research with /research

/research is also available in Telegram and Zalo. The agent runs a repeatable process:

1. Frames the question. It restates what it is investigating and the scope.
2. Shows a plan. You confirm the plan before the research starts.
3. Researches in rounds. Parallel helpers gather and read sources (8 at deep depth).
4. Cross-checks claims. It rates each key claim against the sources it found.
5. Red-teams the draft. Four critic roles look for weak reasoning, missing evidence, and counterarguments.
6. Writes the report. A cited Markdown report and a self-contained HTML report are saved in a folder of their own under research/.

Choose quick, standard, or deep before the run. Quick is for a narrow question; standard balances breadth and time; deep spends more helpers and rounds on contested topics.

What a report contains

• The question and scope, as confirmed in the plan.
• Findings organized by sub-question.
• A rating for each key claim, with the source behind it.
• A red-team section with the strongest counterarguments.
• A source list and links.

Research from a chat channel

Send /research as a message to the paired Telegram or Zalo bot. The agent replies with the plan and asks for confirmation in the chat. Reports are saved on the computer that runs the daemon; the chat gets the summary and file locations.

Choosing between them

You need…	Use
A cited investigation with sources and critique	/research
A long build, edit, or audit that must finish	/goal
A quick answer from one document	Just ask
Work that should run on a schedule	A scheduled task

Related: Chat with documents, Scheduled tasks, Remote access and channels.