AI chat & agents

Every idea comes with an AI assistant that has read all of your research. Ask it questions in plain language — “What’s the strongest evidence for our pricing?”, “Summarize the competitor weaknesses” — and it answers from your idea’s actual data, not generic knowledge.

A chat that knows your idea

The chat panel on your idea’s page keeps full conversation history — every question and answer is saved, and you can return to a conversation weeks later and continue it. The assistant has the context of your idea’s research built in, so follow-up questions make sense without re-explaining everything.

Ask it to search the web

Some questions can’t be answered from your research alone: “What did our main competitor announce last quarter?”, “What’s the current market size for X?”. When the assistant decides it needs current information, it searches the web — you’ll see a “searching the web” indicator, and the answer arrives with source links you can click. It’s your research assistant plus a live browser, in one place.

Let it act — with your permission

Beyond answering questions, the assistant can take actions for you. It can:

• List your ideas and pull up details on any of them
• Summarize your research runs
• Create a new idea in your workspace
• Update the current idea’s name, description, or status
• Add a note to your idea
• Kick off a research phase

Here’s the important part: the assistant never changes anything without asking you first. It separates its tools into two kinds:

• Read tools (looking things up, summarizing) run immediately — they can’t hurt anything, and the results feed the conversation.
• Write tools (creating, updating, adding, spending credits on research) are proposed first. The assistant tells you exactly what it wants to do and pauses. You confirm or deny, and only on your confirmation does it act.

If you deny an action, it doesn’t happen — the assistant acknowledges and moves on. Actions that cost credits, like starting a new research phase, require your explicit confirmation before anything is queued.

Planner mode

For bigger jobs, switch the assistant into planner mode. It lays out a step-by-step plan of the work, runs the read-only steps, and returns to you for confirmation at each point where it needs to change something — so multi-step tasks stay under your control from start to finish.

Enrich your idea with what you learn

Good answers shouldn’t evaporate. When the assistant produces an insight you want to keep, you can save the conversation or pull key findings into your idea’s research — so the value of the conversation becomes part of your permanent record, visible everywhere else in the platform.

Suggested first conversation

1. Ask “What’s the biggest risk to this idea?” and read the answer against your own view.
2. Follow up with “What would strengthen our pitch?” — it draws on your research.
3. Try an action: “Add a note summarizing the competitor launch” — and confirm when it asks.
4. Try planner mode on a small task to see how confirmations work for multi-step work.

Tip

The chat is a thinking partner and an operator, not a decision-maker. Use it to test arguments (“steel-man our competitors’ best case”), but always confirm before it spends credits or writes to your idea.